Canonical read-only response schemas for health and readiness, resolved configuration, opt-in model-state diagnostics, model discovery, decode statistics, recent requests, and persistent teacher-logit cache statistics. Kiln does not authenticate these endpoints. GET /v1/stats/recent-requests can return stored prompt and completion text, so keep the server on loopback or behind an authenticated reverse proxy.
Schema identity
| Property | Value |
|---|---|
| Title | Kiln Read-only Serving and Observability API v1 |
$id |
https://ericflo.github.io/kiln/contracts/kiln-observability-v1.schema.json |
| Dialect | https://json-schema.org/draft/2020-12/schema |
| Root type | Union of 10 public entrypoints |
| Root object | Defined by the selected entrypoint |
Entrypoints
Choose the request or response shape for the operation you are implementing. Each name links to its complete field table and constraints.
| Entrypoint | Purpose | Shape | Unknown fields |
|---|---|---|---|
CacheStatsResponse |
Persistent logit-cache root and occupancy report. | 2 fields | Unknown fields rejected |
ConfigResponse |
Complete resolved configuration and live capacity response. | 21 fields | Unknown fields rejected |
DebugDisabledResponse |
HTTP 403 body returned when model-state diagnostics are not enabled. | 2 fields | Unknown fields rejected |
DebugProvenanceErrorResponse |
HTTP 500 body returned when the resident execution provenance fails validation. | 2 fields | Unknown fields rejected |
DecodeStatsSnapshot |
Rolling request-local decode throughput, ITL tail, and bounded stall diagnosis. | 12 fields | Unknown fields rejected |
HealthResponse |
Complete readiness and runtime diagnostic response returned with HTTP 200 or 503. | 37 fields | Unknown fields rejected |
ModelStateResponse |
Complete opt-in model, adapter, provenance, typed runtime, and cache debug state. | 19 fields | Unknown fields rejected |
ModelsResponse |
OpenAI-compatible list of served models. | 2 fields | Unknown fields rejected |
RequestRecord |
One newest-first bounded recent-request record. The record can contain full prompt and completion text and therefore belongs inside the server's trusted access boundary. | 26 fields | Unknown fields rejected |
Vec_RequestRecord |
Newest-first snapshot of the bounded recent-request ring. | 0 fields | Varies by composed variant |
Composition and conditional rules
Show exact composition rules
{
"oneOf": [
{
"$ref": "#/$defs/CacheStatsResponse"
},
{
"$ref": "#/$defs/ConfigResponse"
},
{
"$ref": "#/$defs/DebugDisabledResponse"
},
{
"$ref": "#/$defs/DebugProvenanceErrorResponse"
},
{
"$ref": "#/$defs/DecodeStatsSnapshot"
},
{
"$ref": "#/$defs/HealthResponse"
},
{
"$ref": "#/$defs/ModelStateResponse"
},
{
"$ref": "#/$defs/ModelsResponse"
},
{
"$ref": "#/$defs/RequestRecord"
},
{
"$ref": "#/$defs/Vec_RequestRecord"
}
]
}
Kiln contract annotations
Show exact Kiln annotations and examples
{
"x-kiln-entrypoints": [
"CacheStatsResponse",
"ConfigResponse",
"DebugDisabledResponse",
"DebugProvenanceErrorResponse",
"DecodeStatsSnapshot",
"HealthResponse",
"ModelStateResponse",
"ModelsResponse",
"RequestRecord",
"Vec_RequestRecord"
],
"x-kiln-examples": {
"CacheStatsResponse": [
{
"root": "/var/lib/kiln/logit-cache",
"stats": {
"total_entries": 12,
"total_bytes": 4096,
"per_teacher": {
"teacher-revision": 12
}
}
}
],
"ConfigResponse": [
{
"effective_configuration": {
"schema_id": "kiln.effective-configuration.v1",
"schema_version": 1,
"effective_config_hash": "sha256:0000000000000000000000000000000000000000000000000000000000000000",
"fixed_field_count": 118,
"dynamic_field_count": 0,
"all_fields_restart_required_to_change": true,
"fields": {}
},
"serving_profile": {
"profile": "stable",
"source": "default",
"immutable_after_startup": false,
"request_overrides_allowed": false,
"effective_policy_source": "value",
"effective_policy": {
"inference_admission": false,
"training_gpu_ownership": false,
"adapter_weight_transitions": false,
"dynamic_kv_resize": false,
"allocator_reclaim": false,
"live_graph_capture": false,
"exclusive_gpu_behavior": "value"
}
},
"accelerator_runtime": {
"schema_id": "value",
"version": 0,
"vulkan_kernel_policy_schema_id": "value",
"vulkan_device_policy_schema_id": "value",
"serving_profile": "stable",
"serving_profile_source": "default",
"kt_api_mode": {
"configured": "auto",
"effective": "auto",
"source": "default"
},
"full_attention_score_budget_mib": {
"configured": 0,
"effective": 0,
"source": "default"
},
"vulkan_device_index": {
"configured": 0,
"effective": 0,
"source": "default"
},
"vulkan_validation": {
"configured": false,
"effective": false,
"source": "default"
},
"cuda_kernel_profile": {
"configured": "native_default",
"effective": "native_default",
"source": "default"
},
"cuda_marlin_profile": {
"configured": "disabled",
"effective": "disabled",
"source": "default"
},
"cuda_flash_backward_mode": {
"configured": "fast",
"effective": "fast",
"source": "default"
},
"metal_kernel_profile": {
"configured": "native_default",
"effective": "native_default",
"source": "default"
},
"rocm_synchronization_mode": {
"configured": "legacy_host_barriers",
"effective": "legacy_host_barriers",
"source": "default"
},
"rocm_strided_batched_matmul_mode": {
"configured": "disabled",
"effective": "disabled",
"source": "default"
},
"rocm_bf16_matmul_output_mode": {
"configured": "f32_then_cast",
"effective": "f32_then_cast",
"source": "default"
},
"rocm_kernel_profile": {
"configured": "native_default",
"effective": "native_default",
"source": "default"
},
"rocm_graph_mode": {
"configured": "profile",
"effective": "lazy_capture_replay",
"source": "default"
},
"rocm_graph_cache_entries": {
"configured": 0,
"effective": 0,
"source": "default"
},
"rocm_graph_cache_max_bytes": {
"configured": 0,
"effective": 0,
"source": "default"
}
},
"rocm_graphs": {
"requested": false,
"capture_requested": false,
"enabled": false,
"capture_enabled": false,
"retained_bytes_accounting_complete": false,
"max_cached_graphs": 0,
"max_retained_bytes": 0,
"capture_attempts": 0,
"capture_successes": 0,
"capture_deferrals": 0,
"capture_failures": 0,
"replay_attempts": 0,
"replay_successes": 0,
"replay_failures": 0,
"failures": 0,
"decode_owner_release_count": 0,
"decode_owner_graph_release_count": 0,
"graph_slot_create_count": 0,
"graph_slot_reuse_count": 0,
"cache_admission_successes": 0,
"cache_evictions": 0,
"cache_evicted_bytes": 0,
"budget_evictions": 0,
"pressure_evictions": 0,
"invalidation_evictions": 0,
"recovery_evictions": 0,
"entry_capacity_rejections": 0,
"byte_budget_rejections": 0,
"accounting_incomplete_rejections": 0,
"pre_capture_entry_capacity_skips": 0,
"pre_capture_byte_budget_skips": 0,
"pre_capture_accounting_incomplete_skips": 0,
"pre_capture_memory_reservation_denied_skips": 0,
"memory_governor_selector_mismatch_skips": 0,
"captured_graph_count": 0,
"graph_slot_count": 0,
"active_graph_slot_count": 0,
"idle_graph_slot_count": 0,
"tracked_decode_owner_count": 0,
"retained_stable_io_bytes": 0,
"retained_capture_arena_bytes": 0,
"retained_blaslt_workspace_bytes": 0,
"retained_slot_state_bytes": 0,
"retained_bytes": 0,
"peak_retained_bytes": 0,
"opaque_native_object_count": 0,
"quarantined_retained_bytes": 0,
"last_transient_candidate_bytes": 0,
"peak_transient_candidate_bytes": 0,
"pre_candidate_headroom_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"candidate_warm_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"pre_native_reservation_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"native_capture_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"native_replay_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"rejected_candidate_cleanup_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"fallbacks": {
"total": 0,
"multi_row_batch_unsupported": 0,
"cold_cache_host_round_trip": 0,
"persistent_host_round_trip": 0,
"shape_dependent_attention": 0,
"graph_cache_capacity": 0,
"graph_cache_byte_budget": 0,
"graph_accounting_incomplete": 0,
"moderate_memory_pressure": 0,
"tight_memory_pressure": 0,
"critical_memory_pressure": 0,
"memory_reservation_denied": 0,
"memory_governor_selector_mismatch": 0,
"capture_failure": 0,
"replay_failure": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
}
},
"rocm_graphs_unavailable_reason": "backend_without_graph_runner",
"rocm_graph_telemetry": {
"current_phase": "pre_candidate_headroom",
"current_phase_elapsed_micros": 0,
"pre_candidate_headroom_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"candidate_warm_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"pre_native_reservation_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"native_capture_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"native_replay_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"rejected_candidate_cleanup_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"last_transient_candidate_bytes": 0,
"peak_transient_candidate_bytes": 0
},
"rocm_graph_telemetry_unavailable_reason": "backend_without_graph_runner",
"cuda_graphs": {
"requested": false,
"capture_allowed_by_serving_profile": true,
"effective_policy_enabled": false,
"max_cached_graphs": 1,
"stable_paged_metadata": true,
"batched_capture_available": false,
"restart_required_to_change": true
},
"decode_runtime": {
"deterministic": {
"enabled": false,
"source": "default"
},
"max_decode_batch": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
}
},
"batching": {
"configuration": {
"rowwise_decode": {
"enabled": false,
"source": "default"
},
"prefix_aware_admission": {
"enabled": false,
"source": "default"
},
"prefill_admission_quantum": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "default"
},
"actor_cycle_idle": {
"milliseconds": 0,
"source": "default",
"enabled": false,
"command_poll_milliseconds": 1
},
"burst_prefill_admission": false,
"actor_prefill_tile_alignment_required": false
},
"actor_active": false
},
"prefix_cache": {
"configuration": {
"enabled": false,
"max_blocks": 1,
"max_entries": 1
},
"effective_enabled": false,
"effective_reason": "value",
"effective_max_blocks": 0,
"effective_max_entries": 0,
"effective_max_state_bytes": 0
},
"streaming_prefill": {
"dispatch": {
"configured_mode": "auto",
"configured_source": "default",
"backend_policy": {
"policy": "never"
},
"effective": {
"policy": "never"
},
"effective_source": "backend_policy"
},
"threshold_tokens": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective_for_auto_mode": 0,
"override_applied_to_backend_auto_policy": false
},
"tile_tokens": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"tape_tile_tokens": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"detached_full_attn_tile_tokens": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"detached_full_attn_boundary_tile_tokens": {
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"detached_full_attn_tape_replay_tile_tokens": {
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"last_token_lm_head": {
"configured": false,
"configured_source": "default",
"effective": false,
"effective_source": "backend_policy"
},
"immutable_after_startup": false,
"restart_required_to_change": false
},
"speculative": {
"enabled": false,
"configured_method": "off",
"configured_effective_method": "off",
"serving_effective_method": "off",
"num_speculative_tokens": 0,
"draft_layers": 0,
"configured_policy_immutable_after_startup": false,
"serving_routable": false,
"serving_unavailable_reason": "value",
"draft_token_ceiling": 0,
"backend_mtp": {
"support": "value",
"native": false
}
},
"model_startup": {
"checkpoint_read": {
"configured_mib_per_second": 1,
"rate_limited": false,
"applicable": false,
"not_applicable_reason": "mock_mode",
"phases": [
"snapshot_copy",
"initial_content_verification",
"post_upload_content_verification"
],
"cancellation_poll_milliseconds": 25,
"current_work_quantum_interruptible": false,
"active_during_inference": false,
"restart_required_to_change": true,
"observed": {
"configured_bytes_per_second": 1,
"snapshot_copy": {
"stage": "value",
"logical_bytes_completed": 0,
"logical_bytes_total": 0,
"rate_limited_bytes_completed": 0,
"elapsed_milliseconds": 0,
"paced_milliseconds": 0,
"complete": true
},
"initial_content_verification": {
"stage": "value",
"logical_bytes_completed": 0,
"logical_bytes_total": 0,
"rate_limited_bytes_completed": 0,
"elapsed_milliseconds": 0,
"paced_milliseconds": 0,
"complete": true
},
"post_upload_content_verification": {
"stage": "value",
"logical_bytes_completed": 0,
"logical_bytes_total": 0,
"rate_limited_bytes_completed": 0,
"elapsed_milliseconds": 0,
"paced_milliseconds": 0,
"complete": true
},
"complete": true
}
},
"accelerator_weight_upload": {
"configured_mib_per_second": 1,
"rate_limited": false,
"applicable": false,
"not_applicable_reason": "mock_mode",
"source_byte_accounting": "base_model_source_bytes",
"cancellation_boundary": "reserve_before_base_and_each_layer; base_upload_then_transpose_then_pack_then_final",
"cancellation_poll_milliseconds": 25,
"current_work_quantum_interruptible": false,
"active_during_inference": false,
"restart_required_to_change": true,
"observed": {
"stage": "value",
"configured_bytes_per_second": 1,
"source_bytes_completed": 1,
"source_bytes_total": 1,
"source_bytes_reserved": 1,
"completed_layers": 1,
"total_layers": 1,
"reserved_layers": 1,
"elapsed_milliseconds": 0,
"paced_milliseconds": 0,
"complete": true
}
}
},
"paths": {
"cache_root": "value",
"cache_root_source": "default",
"restart_required_to_change": true
},
"vram": {
"probe_selector": "value",
"unified": false,
"physical_capacity_bytes": 0,
"physical_capacity_gib": 0,
"physical_capacity_source": "value",
"configured_capacity_bytes": 0,
"configured_capacity_gib": 0,
"effective_capacity_bytes": 0,
"effective_capacity_gib": 0,
"effective_capacity_source": "value",
"configured_capacity_clamped": false,
"live": {
"total_bytes": 0,
"used_bytes": 0,
"available_bytes": 0,
"effective_capacity_available_bytes": 0,
"usable_after_governor_floor_bytes": 0,
"soft_reserved_bytes": 0,
"sample_age_ms": 0,
"sample_max_age_ms": 0,
"total_gib": 0,
"used_gib": 0,
"available_gib": 0,
"effective_capacity_available_gib": 0,
"usable_after_governor_floor_gib": 0,
"soft_reserved_gib": 0,
"pressure": "value",
"source": "value",
"sample_stale": false,
"sampler_required": false,
"sampler_running": false,
"sampler_healthy": false,
"raw_observations": {
"probe_failed": false,
"driver_total_bytes": 0,
"driver_used_bytes": 0,
"driver_free_bytes": 0,
"driver_vram_total_bytes": 0,
"driver_vram_used_bytes": 0,
"driver_gtt_total_bytes": 0,
"driver_gtt_used_bytes": 0,
"host_total_bytes": 0,
"host_available_bytes": 0,
"cgroup_limit_bytes": 0,
"cgroup_high_bytes": 0,
"cgroup_current_bytes": 0,
"cgroup_remaining_bytes": 0,
"unified_reserve_bytes": 0,
"host_backed": {
"total_bytes": 0,
"used_bytes": 0,
"free_bytes": 0,
"total_gib": 0,
"used_gib": 0,
"free_gib": 0
}
}
},
"governor": {
"floor_bytes": 0,
"floor_gib": 0,
"capacity_limit_bytes": 0,
"capacity_limit_gib": 0,
"probe_ms": 0,
"reclaim_mode_requested": "value",
"reclaim_mode_effective": "value",
"reclaim_mode_source": "default",
"reclaim_disabled_by_serving_profile": false
}
},
"kv_cache": {
"num_blocks": 0,
"num_blocks_source": "value",
"fp8_enabled": false,
"autoscaler": {
"requested": false,
"requested_source": "default",
"force_blocks": 0,
"force_blocks_source": "default",
"enabled": false,
"state": "value",
"reason": "value",
"start_blocks": 0,
"min_blocks": 0,
"bytes_per_block": 0
}
},
"training": {
"runtime_device": "value",
"model_weight_device": "value",
"native_training_supported": false,
"optimizer_support": {
"schema": {
"id": "kiln.training-optimizer-support",
"version": 1
},
"backend": "value",
"device": "value",
"base_weight_dtype": "value",
"resolved_lora_parameter_dtype": "value",
"immutable_after_startup": false,
"rounding_modes": [],
"backend_implementation_rounding_modes": [],
"optimizer_tuple_kinds": [],
"workloads": [],
"optimizers": []
},
"checkpoint_policy": {
"mode": "auto"
},
"checkpoint_boundary_policy": {
"recompute_mode": "auto",
"recompute_threshold_tokens": 1,
"anchor_stride": 1,
"cache_target_bytes": 1
},
"checkpoint_segments": 0,
"checkpoint_segments_source": "value",
"checkpointing_enabled": false
},
"memory_budget": {
"total_vram_bytes": 0,
"model_bytes": 0,
"kv_cache_bytes": 0,
"training_budget_bytes": 0,
"total_vram_gib": 0,
"model_gib": 0,
"kv_cache_gib": 0,
"training_budget_gib": 0,
"inference_memory_fraction": 0
},
"generation": {
"default_thinking_enabled": false,
"default_thinking_budget_tokens": 0,
"default_thinking_budget_ms": 0,
"fold_reasoning_into_content": false
},
"operational": {
"bind_host": "value",
"terminal_access": "loopback_only",
"terminal_enabled": false,
"agent_runs_access": "loopback_only",
"agent_runs_enabled": false,
"pi_bin": "value",
"pi_sessions_dir": "value",
"adapter_library_url": "value",
"logit_cache_dir": "value"
}
}
],
"DebugDisabledResponse": [
{
"error": "debug endpoint disabled",
"enable_with": "set server.debug_model_state=true or server.eval_mode=true"
}
],
"DebugProvenanceErrorResponse": [
{
"error": "invalid execution provenance",
"detail": "execution provenance digest mismatch"
}
],
"DecodeStatsSnapshot": [
{
"tok_per_sec": 42.5,
"p50_itl_ms": 23.1,
"p99_itl_ms": 31.8,
"p999_itl_ms": 48.2,
"mean_itl_ms": 24,
"max_itl_ms": 51,
"stall_threshold_ms": 250,
"stall_count": 0,
"unexplained_stall_count": 0,
"stall_reasons": {
"actor_queue": 0,
"actor_admission": 0,
"actor_prefill": 0,
"actor_decode": 0,
"actor_cycle_idle": 0,
"response_delivery": 0,
"handler_queue": 0,
"client_delivery": 0,
"unexplained": 0,
"sampling": 0,
"readback": 0,
"gpu_lock_wait": 0,
"graph_capture": 0,
"graph_replay": 0,
"synchronization": 0,
"resize": 0,
"trim": 0,
"adapter": 0,
"training": 0
},
"sample_count": 84,
"window_secs": 60
}
],
"HealthResponse": [
{
"status": "ok",
"version": "0.1.0",
"uptime_seconds": 0,
"model": "Qwen3.5-4B (32L, 32H, 8KV)",
"backend": "mock",
"backend_runtime": {
"healthy": true,
"quarantined": false,
"reason": "value",
"restart_required": false,
"external_yield_sync": []
},
"serving_profile": {
"profile": "stable",
"source": "default",
"immutable_after_startup": true,
"request_overrides_allowed": false,
"effective_policy_source": "value",
"effective_policy": {
"inference_admission": true,
"training_gpu_ownership": true,
"adapter_weight_transitions": true,
"dynamic_kv_resize": true,
"allocator_reclaim": true,
"live_graph_capture": true,
"exclusive_gpu_behavior": "writer_priority"
}
},
"http": {
"send_buffer_requested_bytes": 0,
"send_buffer_kernel_readback_bytes": 0,
"send_buffer_effective_bytes": 0
},
"model_defaults_profile": {
"name": "Qwen3.5-4B",
"canonical_model_id": "Qwen/Qwen3.5-4B",
"canonical_served_model_id": "Qwen3.5-4B",
"server_default_thinking_enabled": false,
"template_default_thinking_enabled": false,
"eval_mode_default_thinking_enabled": false,
"adapter_dir_policy": "value",
"chat_template_policy": "value",
"supports_enable_thinking_kwarg": false,
"supports_tool_chat_template": false
},
"eval_mode": false,
"debug_model_state": false,
"default_thinking_enabled": false,
"default_thinking_budget_tokens": 0,
"default_thinking_budget_ms": 0,
"fold_reasoning_into_content": false,
"config_hashes": {},
"base_weight_identity": {
"manifest_type": "value",
"aggregate_algorithm": "value",
"aggregate_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000",
"shard_count": 0,
"total_size_bytes": 0
},
"execution_identity": {
"provenance_type": "value",
"backend": "value",
"device": "value",
"inference_dtype": "value",
"training_policy": "value",
"provenance_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000",
"executable_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000",
"numerical_runtime_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000",
"kernel_contract_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000",
"effective_server_config_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000",
"effective_environment_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000"
},
"active_adapter": "value",
"loaded_adapter": "value",
"loaded_adapter_revision": "value",
"loaded_adapter_count": 0,
"adapters_loaded": 0,
"request_count": 0,
"requests": {
"total": 0,
"ok": 0,
"error": 0,
"timeout": 0,
"rejected": 0,
"active": 0,
"active_peak": 0
},
"recent_requests": {
"retained": 0,
"capacity": 0,
"latency_ms": {
"p50": 0,
"p95": 0,
"p99": 0
},
"tokens_per_second": 0,
"timeout_count": 0,
"error_count": 0,
"last_error": {
"id": "value",
"timestamp_unix_ms": 0,
"finish_reason": "value",
"error": "value",
"duration_ms": 0,
"adapter": "value"
}
},
"scheduler": {
"waiting": 0,
"running": 0,
"blocks_used": 0,
"blocks_free": 0,
"blocks_total": 0
},
"gpu_memory": {
"total_vram_bytes": 0,
"model_bytes": 0,
"estimated_model_bytes": 0,
"post_load_used_bytes": 0,
"peak_prefill_used_bytes": 0,
"kv_cache_bytes": 0,
"training_budget_bytes": 0,
"allocated_bytes": 0,
"reserved_bytes": 0,
"total_vram_gb": 0,
"model_gb": 0,
"estimated_model_gb": 0,
"post_load_used_gb": 0,
"peak_prefill_used_gb": 0,
"kv_cache_gb": 0,
"training_budget_gb": 0,
"allocated_gb": 0,
"reserved_gb": 0,
"inference_memory_fraction": 0,
"live": {
"probe_failed": false,
"total_gb": 0,
"used_gb": 0,
"free_gb": 0,
"available_gb": 0,
"soft_reserved_gb": 0,
"pressure": "value",
"source": "value",
"unified": false,
"sample_age_ms": 0,
"sample_max_age_ms": 0,
"sample_stale": false,
"sampler_required": false,
"sampler_running": false,
"sampler_healthy": false,
"host_backed": {
"total_gb": 0,
"used_gb": 0,
"free_gb": 0
}
}
},
"prefix_cache": {
"enabled": false,
"lookup_hits": 0,
"lookup_misses": 0,
"hit_tokens": 0,
"hit_blocks": 0,
"cached_blocks": 0,
"max_blocks": 0,
"cached_entries": 0,
"max_entries": 0,
"cached_state_bytes": 0,
"max_state_bytes": 0,
"active_leases": 0,
"pending_release_entries": 0,
"block_utilization": 0,
"entry_utilization": 0,
"state_utilization": 0
},
"prompt_caches": {
"rendered_prompt": {
"hits": 0,
"misses": 0,
"entries": 0
},
"prompt_token": {
"hits": 0,
"misses": 0,
"entries": 0
}
},
"decode_runtime": {
"configuration": {
"deterministic": {
"enabled": false,
"source": "default"
},
"max_decode_batch": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
}
},
"accelerator_runtime": {
"schema_id": "value",
"version": 0,
"vulkan_kernel_policy_schema_id": "value",
"vulkan_device_policy_schema_id": "value",
"serving_profile": "stable",
"serving_profile_source": "default",
"kt_api_mode": {
"configured": "auto",
"effective": "auto",
"source": "default"
},
"full_attention_score_budget_mib": {
"configured": 0,
"effective": 0,
"source": "default"
},
"vulkan_device_index": {
"configured": 0,
"effective": 0,
"source": "default"
},
"vulkan_validation": {
"configured": false,
"effective": false,
"source": "default"
},
"cuda_kernel_profile": {
"configured": "native_default",
"effective": "native_default",
"source": "default"
},
"cuda_marlin_profile": {
"configured": "disabled",
"effective": "disabled",
"source": "default"
},
"cuda_flash_backward_mode": {
"configured": "fast",
"effective": "fast",
"source": "default"
},
"metal_kernel_profile": {
"configured": "native_default",
"effective": "native_default",
"source": "default"
},
"rocm_synchronization_mode": {
"configured": "legacy_host_barriers",
"effective": "legacy_host_barriers",
"source": "default"
},
"rocm_strided_batched_matmul_mode": {
"configured": "disabled",
"effective": "disabled",
"source": "default"
},
"rocm_bf16_matmul_output_mode": {
"configured": "f32_then_cast",
"effective": "f32_then_cast",
"source": "default"
},
"rocm_kernel_profile": {
"configured": "native_default",
"effective": "native_default",
"source": "default"
},
"rocm_graph_mode": {
"configured": "profile",
"effective": "lazy_capture_replay",
"source": "default"
},
"rocm_graph_cache_entries": {
"configured": 0,
"effective": 0,
"source": "default"
},
"rocm_graph_cache_max_bytes": {
"configured": 0,
"effective": 0,
"source": "default"
}
},
"cuda_synchronization": {
"active": false,
"telemetry_available": false,
"total_device_wait_count": 0,
"total_stream_wait_count": 0,
"total_failure_count": 0,
"total_waited_ns": 0,
"reasons": []
},
"rocm_synchronization": {
"active": false,
"telemetry_available": false,
"cleanup_quarantined": false,
"total_device_wait_count": 0,
"total_stream_wait_count": 0,
"total_waited_ns": 0,
"total_skipped_count": 0,
"reasons": []
},
"batching_configuration": {
"rowwise_decode": {
"enabled": false,
"source": "default"
},
"prefix_aware_admission": {
"enabled": false,
"source": "default"
},
"prefill_admission_quantum": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "default"
},
"actor_cycle_idle": {
"milliseconds": 0,
"source": "default",
"enabled": false,
"command_poll_milliseconds": 1
},
"burst_prefill_admission": false,
"actor_prefill_tile_alignment_required": false
},
"cuda_graphs": {
"requested": false,
"capture_allowed_by_serving_profile": true,
"enabled": false,
"state": "enabled",
"max_cached_graphs": 1,
"stable_paged_metadata": true,
"batched_capture_available": false,
"restart_required_to_change": true
},
"rocm_graphs": {
"requested": false,
"capture_requested": false,
"enabled": false,
"capture_enabled": false,
"state": "enabled",
"unavailable_reason": "backend_without_graph_runner",
"phase_telemetry_available": false,
"phase_telemetry_unavailable_reason": "backend_without_graph_runner",
"current_phase": "pre_candidate_headroom",
"current_phase_elapsed_micros": 0,
"max_cached_graphs": 0,
"max_retained_bytes": 0,
"capture_attempts": 0,
"capture_successes": 0,
"capture_deferrals": 0,
"capture_failures": 0,
"replay_attempts": 0,
"replay_successes": 0,
"replay_failures": 0,
"failures": 0,
"decode_owner_release_count": 0,
"decode_owner_graph_release_count": 0,
"graph_slot_create_count": 0,
"graph_slot_reuse_count": 0,
"cache_admission_successes": 0,
"cache_evictions": 0,
"cache_evicted_bytes": 0,
"budget_evictions": 0,
"pressure_evictions": 0,
"invalidation_evictions": 0,
"recovery_evictions": 0,
"entry_capacity_rejections": 0,
"byte_budget_rejections": 0,
"accounting_incomplete_rejections": 0,
"pre_capture_entry_capacity_skips": 0,
"pre_capture_byte_budget_skips": 0,
"pre_capture_accounting_incomplete_skips": 0,
"pre_capture_memory_reservation_denied_skips": 0,
"memory_governor_selector_mismatch_skips": 0,
"captured_graph_count": 0,
"graph_slot_count": 0,
"active_graph_slot_count": 0,
"idle_graph_slot_count": 0,
"tracked_decode_owner_count": 0,
"retained_stable_io_bytes": 0,
"retained_capture_arena_bytes": 0,
"retained_blaslt_workspace_bytes": 0,
"retained_slot_state_bytes": 0,
"retained_bytes": 0,
"peak_retained_bytes": 0,
"opaque_native_object_count": 0,
"quarantined_retained_bytes": 0,
"last_transient_candidate_bytes": 0,
"peak_transient_candidate_bytes": 0,
"retained_bytes_accounting_complete": false,
"pre_candidate_headroom_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"candidate_warm_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"pre_native_reservation_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"native_capture_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"native_replay_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"rejected_candidate_cleanup_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"fallbacks": {
"total": 0,
"multi_row_batch_unsupported": 0,
"cold_cache_host_round_trip": 0,
"persistent_host_round_trip": 0,
"shape_dependent_attention": 0,
"graph_cache_capacity": 0,
"graph_cache_byte_budget": 0,
"graph_accounting_incomplete": 0,
"moderate_memory_pressure": 0,
"tight_memory_pressure": 0,
"critical_memory_pressure": 0,
"memory_reservation_denied": 0,
"memory_governor_selector_mismatch": 0,
"capture_failure": 0,
"replay_failure": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
}
},
"metal_graphs": {
"enabled": false,
"state": "enabled"
},
"kv_autoscaler": {
"requested": false,
"requested_source": "default",
"force_blocks": 0,
"force_blocks_source": "default",
"enabled": false,
"state": "value",
"reason": "value",
"start_blocks": 0,
"min_blocks": 0,
"bytes_per_block": 0
},
"memory_governor": {
"sample_age_ms": 0,
"sample_max_age_ms": 0,
"automatic_attempts": 0,
"automatic_successful_attempts": 0,
"automatic_zero_yield_attempts": 0,
"automatic_suppressed_attempts": 0,
"automatic_reclaimed_bytes": 0,
"automatic_last_target_bytes": 0,
"automatic_last_reclaimed_bytes": 0,
"automatic_last_duration_us": 0,
"automatic_retry_after_ms": 0,
"automatic_zero_yield_streak": 0,
"reclaim_mode": "value",
"requested_reclaim_mode": "value",
"automatic_monitor_enabled": false,
"sampler_required": false,
"sampler_running": false,
"sampler_healthy": false,
"sample_stale": false,
"source": "default",
"disabled_by_serving_profile": false
},
"batching_engine": {
"snapshot_age_ms": 0,
"stream_stall_grace_ms": 0,
"actor_cycle_idle_ms": 0,
"actor_cycle_idle_count": 0,
"queue_depth": 0,
"active_decode": 0,
"active_prefill": 0,
"active_resident_prefill": 0,
"max_batch_tokens": 0,
"max_prefill_tokens_per_cycle": 0,
"max_prefill_layers_per_cycle": 0,
"max_prefill_admission_quantum": 0,
"max_prefill_staging_slots": 0,
"max_active_requests": 0,
"max_prefill_staging_priority_burst": 0,
"max_decode_batch": 0,
"active_staged_requests": 0,
"max_observed_active_requests": 0,
"current_batch_size": 0,
"last_batch_size": 0,
"max_observed_batch_size": 0,
"slow_decode_forward_count": 0,
"slow_prefill_forward_count": 0,
"last_prefill_tokens": 0,
"last_prefill_layers": 0,
"total_admission_calls": 0,
"slow_admission_count": 0,
"total_decode_forwards": 0,
"total_batched_decode_forwards": 0,
"total_decode_rows": 0,
"total_prefill_admission_cycles": 0,
"total_prefill_forwards": 0,
"total_resident_prefill_attempts": 0,
"total_resident_prefill_forwards": 0,
"total_resident_prefill_initial_declines": 0,
"total_resident_prefill_route_failures": 0,
"total_resident_prefill_rows": 0,
"total_resident_prefill_completed_rows": 0,
"last_resident_prefill_batch_size": 0,
"max_resident_prefill_batch_size": 0,
"total_decode_tokens": 0,
"total_prefill_tokens": 0,
"total_prefill_layers": 0,
"total_prefill_layer_yields": 0,
"total_short_prefill_priority_forwards": 0,
"total_prefill_staging_priority_forwards": 0,
"total_prefill_staging_admissions": 0,
"total_errors": 0,
"response_delivery_in_flight": 0,
"response_delivery_backpressured": 0,
"response_delivery_pending_terminal": 0,
"response_backpressure_events": 0,
"response_backpressure_wait_ms": 0,
"response_stall_evictions": 0,
"response_channel_closed": 0,
"adapter_groups_waiting": 0,
"prefix_deferred_waiting": 0,
"prefix_admission_deferrals": 0,
"last_forward_ms": 0,
"max_decode_forward_ms": 0,
"total_decode_forward_ms": 0,
"last_prefill_ms": 0,
"max_prefill_forward_ms": 0,
"total_prefill_forward_ms": 0,
"last_admission_ms": 0,
"max_admission_ms": 0,
"total_admission_ms": 0,
"total_actor_cycle_idle_ms": 0,
"max_actor_cycle_idle_ms": 0,
"stream_stall_grace_source": "default",
"actor_cycle_idle_source": "default",
"actor_cycle_idle_active": false,
"actor_barrier_adapter_active": false,
"actor_barrier_resize_active": false,
"accepting": false,
"max_batch_tokens_source": "default",
"max_prefill_tokens_per_cycle_source": "default",
"max_prefill_layers_per_cycle_source": "default"
}
},
"prefill_runtime": {
"streaming_prefill": {
"dispatch": {
"configured_mode": "auto",
"configured_source": "default",
"backend_policy": {
"policy": "never"
},
"effective": {
"policy": "never"
},
"effective_source": "backend_policy"
},
"threshold_tokens": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective_for_auto_mode": 0,
"override_applied_to_backend_auto_policy": false
},
"tile_tokens": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"tape_tile_tokens": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"detached_full_attn_tile_tokens": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"detached_full_attn_boundary_tile_tokens": {
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"detached_full_attn_tape_replay_tile_tokens": {
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"last_token_lm_head": {
"configured": false,
"configured_source": "default",
"effective": false,
"effective_source": "backend_policy"
},
"immutable_after_startup": false,
"restart_required_to_change": false
}
},
"training": {
"active_job": {
"job_id": "value",
"progress": 0
},
"queued": 0,
"checkpoint_boundary_policy": {
"recompute_mode": "auto",
"recompute_threshold_tokens": 1,
"anchor_stride": 1,
"cache_target_bytes": 1
}
},
"checks": [
{
"name": "model_loaded",
"pass": true
},
{
"name": "scheduler_responsive",
"pass": true
},
{
"name": "backend_runtime_healthy",
"pass": true
},
{
"name": "inference_admission",
"pass": true
},
{
"name": "inference_prewarm_complete",
"pass": true
},
{
"name": "execution_provenance_valid",
"pass": true
}
],
"vulkan_buffers": {
"live_device_local_buffers": 96,
"live_device_local_bytes": 8589934592,
"live_host_visible_buffers": 24,
"live_host_visible_bytes": 67108864,
"peak_live_bytes": 8724152320,
"device_local_allocations": 140,
"device_local_allocated_bytes": 9663676416,
"device_local_frees": 44,
"device_local_freed_bytes": 1073741824,
"host_visible_allocations": 88,
"host_visible_allocated_bytes": 201326592,
"host_visible_frees": 64,
"host_visible_freed_bytes": 134217728
},
"vulkan_buffer_pool": {
"max_retained_bytes": 3221225472,
"bucket_count": 48,
"buffer_count": 180,
"retained_bytes": 2148204544,
"free_buffer_count": 176,
"free_bytes": 2080768000,
"borrowed_buffer_count": 4,
"borrowed_bytes": 67436544,
"cache_hits": 981000,
"cache_misses": 180,
"eviction_count": 0,
"evicted_bytes": 0,
"uncached_allocation_count": 0,
"uncached_allocated_bytes": 0
}
}
],
"ModelStateResponse": [
{
"model": {
"path": "value",
"served_model_id": "value",
"defaults_profile": {
"name": "value",
"canonical_model_id": "value",
"canonical_served_model_id": "value",
"server_default_thinking_enabled": false,
"template_default_thinking_enabled": false,
"eval_mode_default_thinking_enabled": false,
"adapter_dir_policy": "value",
"chat_template_policy": "value",
"supports_enable_thinking_kwarg": false,
"supports_tool_chat_template": false
},
"num_layers": 0,
"num_attention_heads": 0,
"num_kv_heads": 0,
"max_position_embeddings": 0,
"base_weight_shard_manifest": {
"schema_version": 1,
"manifest_type": "kiln.base-weight-shards.v1",
"aggregate_algorithm": "kiln.base-model-content.v1",
"aggregate_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000",
"total_size_bytes": 1,
"shards": [
{
"filename": "value",
"size_bytes": 1,
"sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000"
}
]
},
"execution_provenance": {
"schema_version": 1,
"provenance_type": "kiln.execution-provenance.v1",
"backend": {
"name": "value",
"device": "value",
"numerical_runtime_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000"
},
"build": {
"package_version": "value",
"target": "value",
"executable_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000"
},
"model": {
"model_config_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000",
"tokenizer_vocab_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000",
"tokenizer_config_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000"
},
"precision": {
"inference_dtype": "value",
"training_policy": "value"
},
"kernels": {
"contract_type": "kiln.kernel-contract.v1",
"versions": {},
"compiled_features": [],
"contract_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000"
},
"configuration": {
"effective_server_config_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000",
"effective_environment_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000"
},
"provenance_sha256": "sha256:0000000000000000000000000000000000000000000000000000000000000000"
},
"backend_runtime": {
"healthy": false,
"quarantined": false,
"reason": "value",
"restart_required": false,
"external_yield_sync": []
}
},
"adapters": {
"adapter_dir": "value",
"active_adapter": "value",
"loaded_adapter": "value",
"loaded_adapter_revision": "value",
"loaded_adapters": [],
"available_adapter_count": 0,
"load_errors": {}
},
"config_hashes": {},
"http": {
"send_buffer_requested_bytes": 0,
"send_buffer_kernel_readback_bytes": 0,
"send_buffer_effective_bytes": 0
},
"decode_runtime": {
"deterministic": {
"enabled": false,
"source": "default"
},
"max_decode_batch": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
}
},
"accelerator_runtime": {
"schema_id": "value",
"version": 0,
"vulkan_kernel_policy_schema_id": "value",
"vulkan_device_policy_schema_id": "value",
"serving_profile": "stable",
"serving_profile_source": "default",
"kt_api_mode": {
"configured": "auto",
"effective": "auto",
"source": "default"
},
"full_attention_score_budget_mib": {
"configured": 0,
"effective": 0,
"source": "default"
},
"vulkan_device_index": {
"configured": 0,
"effective": 0,
"source": "default"
},
"vulkan_validation": {
"configured": false,
"effective": false,
"source": "default"
},
"cuda_kernel_profile": {
"configured": "native_default",
"effective": "native_default",
"source": "default"
},
"cuda_marlin_profile": {
"configured": "disabled",
"effective": "disabled",
"source": "default"
},
"cuda_flash_backward_mode": {
"configured": "fast",
"effective": "fast",
"source": "default"
},
"metal_kernel_profile": {
"configured": "native_default",
"effective": "native_default",
"source": "default"
},
"rocm_synchronization_mode": {
"configured": "legacy_host_barriers",
"effective": "legacy_host_barriers",
"source": "default"
},
"rocm_strided_batched_matmul_mode": {
"configured": "disabled",
"effective": "disabled",
"source": "default"
},
"rocm_bf16_matmul_output_mode": {
"configured": "f32_then_cast",
"effective": "f32_then_cast",
"source": "default"
},
"rocm_kernel_profile": {
"configured": "native_default",
"effective": "native_default",
"source": "default"
},
"rocm_graph_mode": {
"configured": "profile",
"effective": "lazy_capture_replay",
"source": "default"
},
"rocm_graph_cache_entries": {
"configured": 0,
"effective": 0,
"source": "default"
},
"rocm_graph_cache_max_bytes": {
"configured": 0,
"effective": 0,
"source": "default"
}
},
"cuda_graphs": {
"requested": false,
"capture_allowed_by_serving_profile": true,
"effective_policy_enabled": false,
"max_cached_graphs": 1,
"stable_paged_metadata": true,
"batched_capture_available": false,
"restart_required_to_change": true
},
"cuda_synchronization": {
"active": false,
"telemetry_available": false,
"total_device_wait_count": 0,
"total_stream_wait_count": 0,
"total_failure_count": 0,
"total_waited_ns": 0,
"reasons": []
},
"rocm_synchronization": {
"active": false,
"telemetry_available": false,
"cleanup_quarantined": false,
"total_device_wait_count": 0,
"total_stream_wait_count": 0,
"total_waited_ns": 0,
"total_skipped_count": 0,
"reasons": []
},
"rocm_graphs": {
"requested": false,
"capture_requested": false,
"enabled": false,
"capture_enabled": false,
"retained_bytes_accounting_complete": false,
"max_cached_graphs": 0,
"max_retained_bytes": 0,
"capture_attempts": 0,
"capture_successes": 0,
"capture_deferrals": 0,
"capture_failures": 0,
"replay_attempts": 0,
"replay_successes": 0,
"replay_failures": 0,
"failures": 0,
"decode_owner_release_count": 0,
"decode_owner_graph_release_count": 0,
"graph_slot_create_count": 0,
"graph_slot_reuse_count": 0,
"cache_admission_successes": 0,
"cache_evictions": 0,
"cache_evicted_bytes": 0,
"budget_evictions": 0,
"pressure_evictions": 0,
"invalidation_evictions": 0,
"recovery_evictions": 0,
"entry_capacity_rejections": 0,
"byte_budget_rejections": 0,
"accounting_incomplete_rejections": 0,
"pre_capture_entry_capacity_skips": 0,
"pre_capture_byte_budget_skips": 0,
"pre_capture_accounting_incomplete_skips": 0,
"pre_capture_memory_reservation_denied_skips": 0,
"memory_governor_selector_mismatch_skips": 0,
"captured_graph_count": 0,
"graph_slot_count": 0,
"active_graph_slot_count": 0,
"idle_graph_slot_count": 0,
"tracked_decode_owner_count": 0,
"retained_stable_io_bytes": 0,
"retained_capture_arena_bytes": 0,
"retained_blaslt_workspace_bytes": 0,
"retained_slot_state_bytes": 0,
"retained_bytes": 0,
"peak_retained_bytes": 0,
"opaque_native_object_count": 0,
"quarantined_retained_bytes": 0,
"last_transient_candidate_bytes": 0,
"peak_transient_candidate_bytes": 0,
"pre_candidate_headroom_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"candidate_warm_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"pre_native_reservation_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"native_capture_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"native_replay_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"rejected_candidate_cleanup_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"fallbacks": {
"total": 0,
"multi_row_batch_unsupported": 0,
"cold_cache_host_round_trip": 0,
"persistent_host_round_trip": 0,
"shape_dependent_attention": 0,
"graph_cache_capacity": 0,
"graph_cache_byte_budget": 0,
"graph_accounting_incomplete": 0,
"moderate_memory_pressure": 0,
"tight_memory_pressure": 0,
"critical_memory_pressure": 0,
"memory_reservation_denied": 0,
"memory_governor_selector_mismatch": 0,
"capture_failure": 0,
"replay_failure": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
}
},
"rocm_graphs_unavailable_reason": "backend_without_graph_runner",
"rocm_graph_telemetry": {
"current_phase": "pre_candidate_headroom",
"current_phase_elapsed_micros": 0,
"pre_candidate_headroom_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"candidate_warm_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"pre_native_reservation_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"native_capture_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"native_replay_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"rejected_candidate_cleanup_phase": {
"calls": 0,
"slow": 0,
"total_duration_micros": 0,
"max_duration_micros": 0
},
"last_transient_candidate_bytes": 0,
"peak_transient_candidate_bytes": 0
},
"rocm_graph_telemetry_unavailable_reason": "backend_without_graph_runner",
"kv_autoscaler": {
"requested": false,
"requested_source": "default",
"force_blocks": 0,
"force_blocks_source": "default",
"enabled": false,
"state": "value",
"reason": "value",
"start_blocks": 0,
"min_blocks": 0,
"bytes_per_block": 0
},
"streaming_prefill": {
"dispatch": {
"configured_mode": "auto",
"configured_source": "default",
"backend_policy": {
"policy": "never"
},
"effective": {
"policy": "never"
},
"effective_source": "backend_policy"
},
"threshold_tokens": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective_for_auto_mode": 0,
"override_applied_to_backend_auto_policy": false
},
"tile_tokens": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"tape_tile_tokens": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"detached_full_attn_tile_tokens": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"detached_full_attn_boundary_tile_tokens": {
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"detached_full_attn_tape_replay_tile_tokens": {
"backend_policy": 0,
"effective": 0,
"effective_source": "backend_policy"
},
"last_token_lm_head": {
"configured": false,
"configured_source": "default",
"effective": false,
"effective_source": "backend_policy"
},
"immutable_after_startup": false,
"restart_required_to_change": false
},
"training": {
"checkpoint_boundary_policy": {
"recompute_mode": "auto",
"recompute_threshold_tokens": 1,
"anchor_stride": 1,
"cache_target_bytes": 1
}
},
"batching_engine": {
"backend": "value",
"enabled": false,
"configuration": {
"rowwise_decode": {
"enabled": false,
"source": "default"
},
"prefix_aware_admission": {
"enabled": false,
"source": "default"
},
"prefill_admission_quantum": {
"configured": 0,
"configured_source": "default",
"backend_policy": 0,
"effective": 0,
"effective_source": "default"
},
"actor_cycle_idle": {
"milliseconds": 0,
"source": "default",
"enabled": false,
"command_poll_milliseconds": 1
},
"burst_prefill_admission": false,
"actor_prefill_tile_alignment_required": false
},
"snapshot": {
"snapshot_age_ms": 0,
"stream_stall_grace_ms": 0,
"actor_cycle_idle_ms": 0,
"actor_cycle_idle_count": 0,
"queue_depth": 0,
"active_decode": 0,
"active_prefill": 0,
"active_resident_prefill": 0,
"max_batch_tokens": 0,
"max_prefill_tokens_per_cycle": 0,
"max_prefill_layers_per_cycle": 0,
"max_prefill_admission_quantum": 0,
"max_prefill_staging_slots": 0,
"max_active_requests": 0,
"max_prefill_staging_priority_burst": 0,
"max_decode_batch": 0,
"active_staged_requests": 0,
"max_observed_active_requests": 0,
"current_batch_size": 0,
"last_batch_size": 0,
"max_observed_batch_size": 0,
"slow_decode_forward_count": 0,
"slow_prefill_forward_count": 0,
"last_prefill_tokens": 0,
"last_prefill_layers": 0,
"total_admission_calls": 0,
"slow_admission_count": 0,
"total_decode_forwards": 0,
"total_batched_decode_forwards": 0,
"total_decode_rows": 0,
"total_prefill_admission_cycles": 0,
"total_prefill_forwards": 0,
"total_resident_prefill_attempts": 0,
"total_resident_prefill_forwards": 0,
"total_resident_prefill_initial_declines": 0,
"total_resident_prefill_route_failures": 0,
"total_resident_prefill_rows": 0,
"total_resident_prefill_completed_rows": 0,
"last_resident_prefill_batch_size": 0,
"max_resident_prefill_batch_size": 0,
"total_decode_tokens": 0,
"total_prefill_tokens": 0,
"total_prefill_layers": 0,
"total_prefill_layer_yields": 0,
"total_short_prefill_priority_forwards": 0,
"total_prefill_staging_priority_forwards": 0,
"total_prefill_staging_admissions": 0,
"total_errors": 0,
"response_delivery_in_flight": 0,
"response_delivery_backpressured": 0,
"response_delivery_pending_terminal": 0,
"response_backpressure_events": 0,
"response_backpressure_wait_ms": 0,
"response_stall_evictions": 0,
"response_channel_closed": 0,
"adapter_groups_waiting": 0,
"prefix_deferred_waiting": 0,
"prefix_admission_deferrals": 0,
"last_forward_ms": 0,
"max_decode_forward_ms": 0,
"total_decode_forward_ms": 0,
"last_prefill_ms": 0,
"max_prefill_forward_ms": 0,
"total_prefill_forward_ms": 0,
"last_admission_ms": 0,
"max_admission_ms": 0,
"total_admission_ms": 0,
"total_actor_cycle_idle_ms": 0,
"max_actor_cycle_idle_ms": 0,
"stream_stall_grace_source": "default",
"actor_cycle_idle_source": "default",
"actor_cycle_idle_active": false,
"actor_barrier_adapter_active": false,
"actor_barrier_resize_active": false,
"accepting": false,
"max_batch_tokens_source": "default",
"max_prefill_tokens_per_cycle_source": "default",
"max_prefill_layers_per_cycle_source": "default"
}
},
"thinking": {
"eval_mode": false,
"default_thinking_enabled": false,
"default_thinking_budget_tokens": 0,
"default_thinking_budget_ms": 0,
"profile_server_default_thinking_enabled": false,
"template_default_thinking_enabled": false,
"eval_mode_default_thinking_enabled": false
},
"caches": {
"deterministic_completion_entries": 0,
"deterministic_chat_request_entries": 0,
"deterministic_chat_choices_entries": 0,
"deterministic_batch_entries": 0,
"batched_recurrent_state": {
"entry_present": false,
"capacity_rows": 0,
"logical_rows": 0,
"resident": false,
"active_leases": 0,
"max_active_leases": 0,
"take_hit_count": 0,
"take_miss_count": 0,
"take_miss_while_leased_count": 0,
"exact_reuse_count": 0,
"resident_capacity_reuse_count": 0,
"resident_prefix_view_count": 0,
"resident_refresh_count": 0,
"fresh_assembly_count": 0,
"rejected_missing_row_ids_count": 0,
"rejected_nonresident_rows_count": 0,
"rejected_nonresident_cache_count": 0,
"rejected_insufficient_capacity_count": 0,
"park_count": 0,
"park_replacement_eviction_count": 0,
"explicit_invalidation_count": 0,
"explicit_invalidation_eviction_count": 0,
"completed_row_preservation_count": 0,
"completed_row_eviction_count": 0,
"lease_drop_eviction_count": 0,
"resident_prefix_snapshot_suppression_count": 0
},
"resident_recurrent_state": {
"entry_count": 0,
"buffer_bytes": 0,
"allocation_bytes": 0
},
"rendered_prompt": {
"hits": 0,
"misses": 0,
"entries": 0
},
"prompt_token": {
"hits": 0,
"misses": 0,
"entries": 0
},
"prefix_cache": {
"enabled": false,
"lookup_hits": 0,
"lookup_misses": 0,
"hit_tokens": 0,
"hit_blocks": 0,
"cached_blocks": 0,
"max_blocks": 0,
"cached_entries": 0,
"max_entries": 0,
"cached_state_bytes": 0,
"max_state_bytes": 0,
"active_leases": 0,
"pending_release_entries": 0
}
}
}
],
"ModelsResponse": [
{
"object": "list",
"data": [
{
"id": "Qwen3.5-4B",
"object": "model",
"owned_by": "kiln"
}
]
}
],
"RequestRecord": [
{
"id": "chatcmpl-example",
"timestamp_unix_ms": 1800000000000,
"model": "Qwen3.5-4B",
"prompt_preview": "Explain bounded thinking.",
"completion_preview": "A bounded thinking request...",
"prompt_tokens": 12,
"completion_tokens": 48,
"duration_ms": 730,
"streamed": true,
"finish_reason": "stop",
"ttft_ms": 88,
"model_prefill_ms": 41,
"model_decode_ms": 620,
"thinking_mode": "enabled",
"prefix_cache": "miss",
"client": "dashboard",
"thinking_budget": {
"configured": true,
"max_tokens": 32,
"tokens_source": "request",
"time_source": "unlimited",
"applied": true,
"triggered": true,
"trigger": "tokens",
"closed": true,
"thinking_tokens": 32,
"thinking_time_ms": 410
},
"latency": {
"emitted_tokens": 48,
"gap_samples": 47,
"retained_gap_samples": 47,
"gap_samples_truncated": false,
"ttft_ms": 88,
"itl_ms_p50": 12,
"itl_ms_p99": 31.8,
"itl_ms_p999": 35,
"max_itl_ms": 36,
"stall_threshold_ms": 250,
"stall_count": 0,
"unexplained_stall_count": 0,
"stall_reasons": {
"actor_queue": 0,
"actor_admission": 0,
"actor_prefill": 0,
"actor_decode": 0,
"actor_cycle_idle": 0,
"response_delivery": 0,
"handler_queue": 0,
"client_delivery": 0,
"unexplained": 0,
"sampling": 0,
"readback": 0,
"gpu_lock_wait": 0,
"graph_capture": 0,
"graph_replay": 0,
"synchronization": 0,
"resize": 0,
"trim": 0,
"adapter": 0,
"training": 0
},
"phases": {
"actor_queue_ms": 8,
"actor_admission_ms": 1,
"tokenization_ms": 2,
"prefill_ms": 39,
"decode_ms": 580,
"actor_cycle_idle_ms": 0,
"sampling_ms": null,
"readback_ms": null,
"response_delivery_ms": 3,
"handler_queue_ms": 1,
"client_delivery_ms": 2,
"gpu_lock_wait_ms": 0.7,
"graph_capture_ms": null,
"graph_replay_ms": null,
"synchronization_ms": 1.4,
"resize_ms": null,
"trim_ms": null,
"adapter_ms": null,
"training_ms": null,
"unexplained_ms": 1
}
}
}
],
"Vec_RequestRecord": [
[
{
"id": "chatcmpl-example",
"timestamp_unix_ms": 1800000000000,
"model": "Qwen3.5-4B",
"prompt_preview": "Explain bounded thinking.",
"completion_preview": "A bounded thinking request...",
"prompt_tokens": 12,
"completion_tokens": 48,
"duration_ms": 730,
"streamed": true,
"finish_reason": "stop",
"ttft_ms": 88,
"model_prefill_ms": 41,
"model_decode_ms": 620,
"thinking_mode": "enabled",
"prefix_cache": "miss",
"client": "dashboard",
"thinking_budget": {
"configured": true,
"max_tokens": 32,
"tokens_source": "request",
"time_source": "unlimited",
"applied": true,
"triggered": true,
"trigger": "tokens",
"closed": true,
"thinking_tokens": 32,
"thinking_time_ms": 410
},
"latency": {
"emitted_tokens": 48,
"gap_samples": 47,
"retained_gap_samples": 47,
"gap_samples_truncated": false,
"ttft_ms": 88,
"itl_ms_p50": 12,
"itl_ms_p99": 31.8,
"itl_ms_p999": 35,
"max_itl_ms": 36,
"stall_threshold_ms": 250,
"stall_count": 0,
"unexplained_stall_count": 0,
"stall_reasons": {
"actor_queue": 0,
"actor_admission": 0,
"actor_prefill": 0,
"actor_decode": 0,
"actor_cycle_idle": 0,
"response_delivery": 0,
"handler_queue": 0,
"client_delivery": 0,
"unexplained": 0,
"sampling": 0,
"readback": 0,
"gpu_lock_wait": 0,
"graph_capture": 0,
"graph_replay": 0,
"synchronization": 0,
"resize": 0,
"trim": 0,
"adapter": 0,
"training": 0
},
"phases": {
"actor_queue_ms": 8,
"actor_admission_ms": 1,
"tokenization_ms": 2,
"prefill_ms": 39,
"decode_ms": 580,
"actor_cycle_idle_ms": 0,
"sampling_ms": null,
"readback_ms": null,
"response_delivery_ms": 3,
"handler_queue_ms": 1,
"client_delivery_ms": 2,
"gpu_lock_wait_ms": 0.7,
"graph_capture_ms": null,
"graph_replay_ms": null,
"synchronization_ms": 1.4,
"resize_ms": null,
"trim_ms": null,
"adapter_ms": null,
"training_ms": null,
"unexplained_ms": 1
}
}
}
]
]
},
"x-kiln-field-schema-status": "complete",
"x-kiln-health-http-status": {
"200": {
"status": "ok"
},
"503": {
"status": [
"degraded",
"maintenance"
]
}
}
}
Definitions
AcceleratorWeightUploadConfigResponseResolved startup-only accelerator-weight upload policy and completed real-model observation.11 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured_mib_per_second |
yes | PositiveInteger | null |
- | Configured accelerator-upload rate in MiB per second; null disables pacing. |
rate_limited |
yes | Boolean |
- | Rate limited. A serialized Rust boolean. |
applicable |
yes | Boolean |
- | Applicable. A serialized Rust boolean. |
not_applicable_reason |
yes | any | null |
- | Why upload pacing does not apply; null when the policy applies. |
source_byte_accounting |
yes | any |
const "base_model_source_bytes" | Source byte accounting. |
cancellation_boundary |
yes | any |
const "reserve_before_base_and_each_layer; base_upload_then_transpose_then_pack_then_final" | Cancellation boundary. |
cancellation_poll_milliseconds |
yes | any |
const 25 | Cancellation poll milliseconds. |
current_work_quantum_interruptible |
yes | any |
const false | Current work quantum interruptible. |
active_during_inference |
yes | any |
const false | Active during inference. |
restart_required_to_change |
yes | any |
const true | Restart required to change. |
observed |
yes | AcceleratorWeightUploadReport | null |
- | Observed. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "AcceleratorWeightUploadConfigResponse"
}
AcceleratorWeightUploadReportExact completed source-byte, layer, elapsed-time, and pacing accounting for eager base-model accelerator upload.11 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
stage |
yes | NonEmptyString |
- | Stage. A non-empty serialized UTF-8 Rust string. |
configured_bytes_per_second |
yes | PositiveInteger | null |
- | Configured bytes per second. |
source_bytes_completed |
yes | PositiveInteger |
- | Source bytes completed. A positive Rust integer represented exactly as a JSON number. |
source_bytes_total |
yes | PositiveInteger |
- | Source bytes total. A positive Rust integer represented exactly as a JSON number. |
source_bytes_reserved |
yes | PositiveInteger |
- | Source bytes reserved. A positive Rust integer represented exactly as a JSON number. |
completed_layers |
yes | PositiveInteger |
- | Completed layers. A positive Rust integer represented exactly as a JSON number. |
total_layers |
yes | PositiveInteger |
- | Total layers. A positive Rust integer represented exactly as a JSON number. |
reserved_layers |
yes | PositiveInteger |
- | Reserved layers. A positive Rust integer represented exactly as a JSON number. |
elapsed_milliseconds |
yes | NonNegativeInteger |
- | Elapsed milliseconds. A non-negative Rust integer represented exactly as a JSON number. |
paced_milliseconds |
yes | NonNegativeInteger |
- | Paced milliseconds. A non-negative Rust integer represented exactly as a JSON number. |
complete |
yes | any |
const true | Complete. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "kiln_model::AcceleratorWeightUploadReport"
}
ActiveJobInfoThe active training job and its reported progress.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
job_id |
yes | String |
- | Job ID. A serialized UTF-8 Rust string. |
progress |
yes | FiniteNumber |
- | Progress. A finite Rust floating-point value. serde_json rejects non-finite values. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ActiveJobInfo"
}
ActorCycleIdleDiagnosticsConfigured cooperative safe-boundary actor idle and control-command polling contract.4 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
milliseconds |
yes | NonNegativeInteger |
- | Milliseconds. A non-negative Rust integer represented exactly as a JSON number. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
enabled |
yes | Boolean |
- | Enabled. A serialized Rust boolean. |
command_poll_milliseconds |
yes | PositiveInteger |
- | Command poll milliseconds. A positive Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ActorCycleIdleDiagnostics"
}
AdapterDebugStateActive, loaded, discoverable, and failed adapter state.7 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
adapter_dir |
yes | String |
- | Adapter dir. A serialized UTF-8 Rust string. |
active_adapter |
yes | String | null |
- | Active adapter. |
loaded_adapter |
yes | String | null |
- | Loaded adapter. |
loaded_adapter_revision |
yes | String | null |
- | Loaded adapter revision. |
loaded_adapters |
yes | array<LoadedAdapterDebugState> |
- | Loaded adapters. |
available_adapter_count |
yes | NonNegativeInteger |
- | Available adapter count. A non-negative Rust integer represented exactly as a JSON number. |
load_errors |
yes | object |
- | Load errors. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "AdapterDebugState"
}
BackendRuntimeInfoBackend health, quarantine, restart, and external-yield settlement state.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
healthy |
yes | Boolean |
- | Healthy. A serialized Rust boolean. |
quarantined |
yes | Boolean |
- | Quarantined. A serialized Rust boolean. |
reason |
yes | String | null |
- | Reason. |
restart_required |
yes | Boolean |
- | Restart required. A serialized Rust boolean. |
external_yield_sync |
yes | array<ExternalYieldSyncStats> |
- | External yield sync. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "BackendRuntimeInfo | BackendRuntimeDebugState"
}
BaseWeightIdentitySummaryCompact loaded base-weight identity.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
manifest_type |
yes | String |
- | Manifest type. A serialized UTF-8 Rust string. |
aggregate_algorithm |
yes | String |
- | Aggregate algorithm. A serialized UTF-8 Rust string. |
aggregate_sha256 |
yes | Sha256 |
- | Aggregate SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
shard_count |
yes | NonNegativeInteger |
- | Shard count. A non-negative Rust integer represented exactly as a JSON number. |
total_size_bytes |
yes | NonNegativeInteger |
- | Total size bytes. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "BaseWeightIdentitySummary"
}
BaseWeightShardIdentityExact identity of one logical safetensors shard.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
filename |
yes | NonEmptyString |
- | Path-safe logical safetensors shard filename. |
size_bytes |
yes | PositiveInteger |
- | Size bytes. A positive Rust integer represented exactly as a JSON number. |
sha256 |
yes | Sha256 |
- | SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "BaseWeightShardIdentity"
}
BaseWeightShardManifestCanonical portable manifest for the exact loaded base-weight shards.6 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
schema_version |
yes | any |
const 1 | Schema version. |
manifest_type |
yes | any |
const "kiln.base-weight-shards.v1" | Manifest type. |
aggregate_algorithm |
yes | any |
const "kiln.base-model-content.v1" | Aggregate algorithm. |
aggregate_sha256 |
yes | Sha256 |
- | Aggregate SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
total_size_bytes |
yes | PositiveInteger |
- | Total size bytes. A positive Rust integer represented exactly as a JSON number. |
shards |
yes | array<BaseWeightShardIdentity> |
minimum items 1 | Shards. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "BaseWeightShardManifest"
}
BatchedStateCacheStatsCurrent ownership and process-lifetime lifecycle counters for the parked batched recurrent-state cache.26 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
entry_present |
yes | Boolean |
- | Entry present. A serialized Rust boolean. |
capacity_rows |
yes | NonNegativeInteger |
- | Capacity rows. A non-negative Rust integer represented exactly as a JSON number. |
logical_rows |
yes | NonNegativeInteger |
- | Logical rows. A non-negative Rust integer represented exactly as a JSON number. |
resident |
yes | Boolean |
- | Resident. A serialized Rust boolean. |
active_leases |
yes | NonNegativeInteger |
- | Active leases. A non-negative Rust integer represented exactly as a JSON number. |
max_active_leases |
yes | NonNegativeInteger |
- | Max active leases. A non-negative Rust integer represented exactly as a JSON number. |
take_hit_count |
yes | NonNegativeInteger |
- | Take hit count. A non-negative Rust integer represented exactly as a JSON number. |
take_miss_count |
yes | NonNegativeInteger |
- | Take miss count. A non-negative Rust integer represented exactly as a JSON number. |
take_miss_while_leased_count |
yes | NonNegativeInteger |
- | Take miss while leased count. A non-negative Rust integer represented exactly as a JSON number. |
exact_reuse_count |
yes | NonNegativeInteger |
- | Exact reuse count. A non-negative Rust integer represented exactly as a JSON number. |
resident_capacity_reuse_count |
yes | NonNegativeInteger |
- | Resident capacity reuse count. A non-negative Rust integer represented exactly as a JSON number. |
resident_prefix_view_count |
yes | NonNegativeInteger |
- | Resident prefix view count. A non-negative Rust integer represented exactly as a JSON number. |
resident_refresh_count |
yes | NonNegativeInteger |
- | Resident refresh count. A non-negative Rust integer represented exactly as a JSON number. |
fresh_assembly_count |
yes | NonNegativeInteger |
- | Fresh assembly count. A non-negative Rust integer represented exactly as a JSON number. |
rejected_missing_row_ids_count |
yes | NonNegativeInteger |
- | Rejected missing row IDs count. A non-negative Rust integer represented exactly as a JSON number. |
rejected_nonresident_rows_count |
yes | NonNegativeInteger |
- | Rejected nonresident rows count. A non-negative Rust integer represented exactly as a JSON number. |
rejected_nonresident_cache_count |
yes | NonNegativeInteger |
- | Rejected nonresident cache count. A non-negative Rust integer represented exactly as a JSON number. |
rejected_insufficient_capacity_count |
yes | NonNegativeInteger |
- | Rejected insufficient capacity count. A non-negative Rust integer represented exactly as a JSON number. |
park_count |
yes | NonNegativeInteger |
- | Park count. A non-negative Rust integer represented exactly as a JSON number. |
park_replacement_eviction_count |
yes | NonNegativeInteger |
- | Park replacement eviction count. A non-negative Rust integer represented exactly as a JSON number. |
explicit_invalidation_count |
yes | NonNegativeInteger |
- | Explicit invalidation count. A non-negative Rust integer represented exactly as a JSON number. |
explicit_invalidation_eviction_count |
yes | NonNegativeInteger |
- | Explicit invalidation eviction count. A non-negative Rust integer represented exactly as a JSON number. |
completed_row_preservation_count |
yes | NonNegativeInteger |
- | Completed row preservation count. A non-negative Rust integer represented exactly as a JSON number. |
completed_row_eviction_count |
yes | NonNegativeInteger |
- | Completed row eviction count. A non-negative Rust integer represented exactly as a JSON number. |
lease_drop_eviction_count |
yes | NonNegativeInteger |
- | Lease drop eviction count. A non-negative Rust integer represented exactly as a JSON number. |
resident_prefix_snapshot_suppression_count |
yes | NonNegativeInteger |
- | Resident prefix snapshot suppression count. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "BatchedStateCacheStats"
}
BatchingConfigResponseResolved batching policy plus live actor state.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configuration |
yes | BatchingRuntimeConfig |
- | Configuration. Runtime-ready batching policy resolved once after backend selection. |
actor_active |
yes | Boolean |
- | Actor active. A serialized Rust boolean. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "BatchingConfigResponse"
}
BatchingEffectiveSourceThe authority that selected an effective batching value.0 fields
Type: string. Constraints: enum "default", "backend_policy", "config_file", "environment", "command_line", "effective_decode_width".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "BatchingEffectiveSource"
}
BatchingEngineDebugStateSelected batching backend, resolved configuration, and live execution state.4 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
backend |
yes | String |
- | Backend. A serialized UTF-8 Rust string. |
enabled |
yes | Boolean |
- | Enabled. A serialized Rust boolean. |
configuration |
yes | BatchingRuntimeConfig |
- | Configuration. Runtime-ready batching policy resolved once after backend selection. |
snapshot |
yes | BatchingEngineSnapshot | null |
- | Snapshot. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "BatchingEngineDebugState"
}
BatchingEngineSnapshotCached batching-engine admission, execution, latency, backpressure, and fairness counters.78 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
snapshot_age_ms |
yes | NonNegativeInteger |
- | Snapshot age ms. A non-negative Rust integer represented exactly as a JSON number. |
stream_stall_grace_ms |
yes | NonNegativeInteger |
- | Stream stall grace ms. A non-negative Rust integer represented exactly as a JSON number. |
actor_cycle_idle_ms |
yes | NonNegativeInteger |
- | Actor cycle idle ms. A non-negative Rust integer represented exactly as a JSON number. |
actor_cycle_idle_count |
yes | NonNegativeInteger |
- | Actor cycle idle count. A non-negative Rust integer represented exactly as a JSON number. |
queue_depth |
yes | NonNegativeInteger |
- | Queue depth. A non-negative Rust integer represented exactly as a JSON number. |
active_decode |
yes | NonNegativeInteger |
- | Active decode. A non-negative Rust integer represented exactly as a JSON number. |
active_prefill |
yes | NonNegativeInteger |
- | Active prefill. A non-negative Rust integer represented exactly as a JSON number. |
active_resident_prefill |
yes | NonNegativeInteger |
- | Active resident prefill. A non-negative Rust integer represented exactly as a JSON number. |
max_batch_tokens |
yes | NonNegativeInteger |
- | Max batch tokens. A non-negative Rust integer represented exactly as a JSON number. |
max_prefill_tokens_per_cycle |
yes | NonNegativeInteger |
- | Max prefill tokens per cycle. A non-negative Rust integer represented exactly as a JSON number. |
max_prefill_layers_per_cycle |
yes | NonNegativeInteger |
- | Max prefill layers per cycle. A non-negative Rust integer represented exactly as a JSON number. |
max_prefill_admission_quantum |
yes | NonNegativeInteger |
- | Max prefill admission quantum. A non-negative Rust integer represented exactly as a JSON number. |
max_prefill_staging_slots |
yes | NonNegativeInteger |
- | Max prefill staging slots. A non-negative Rust integer represented exactly as a JSON number. |
max_active_requests |
yes | NonNegativeInteger |
- | Max active requests. A non-negative Rust integer represented exactly as a JSON number. |
max_prefill_staging_priority_burst |
yes | NonNegativeInteger |
- | Max prefill staging priority burst. A non-negative Rust integer represented exactly as a JSON number. |
max_decode_batch |
yes | NonNegativeInteger |
- | Max decode batch. A non-negative Rust integer represented exactly as a JSON number. |
active_staged_requests |
yes | NonNegativeInteger |
- | Active staged requests. A non-negative Rust integer represented exactly as a JSON number. |
max_observed_active_requests |
yes | NonNegativeInteger |
- | Max observed active requests. A non-negative Rust integer represented exactly as a JSON number. |
current_batch_size |
yes | NonNegativeInteger |
- | Current batch size. A non-negative Rust integer represented exactly as a JSON number. |
last_batch_size |
yes | NonNegativeInteger |
- | Last batch size. A non-negative Rust integer represented exactly as a JSON number. |
max_observed_batch_size |
yes | NonNegativeInteger |
- | Max observed batch size. A non-negative Rust integer represented exactly as a JSON number. |
slow_decode_forward_count |
yes | NonNegativeInteger |
- | Slow decode forward count. A non-negative Rust integer represented exactly as a JSON number. |
slow_prefill_forward_count |
yes | NonNegativeInteger |
- | Slow prefill forward count. A non-negative Rust integer represented exactly as a JSON number. |
last_prefill_tokens |
yes | NonNegativeInteger |
- | Last prefill tokens. A non-negative Rust integer represented exactly as a JSON number. |
last_prefill_layers |
yes | NonNegativeInteger |
- | Last prefill layers. A non-negative Rust integer represented exactly as a JSON number. |
total_admission_calls |
yes | NonNegativeInteger |
- | Total admission calls. A non-negative Rust integer represented exactly as a JSON number. |
slow_admission_count |
yes | NonNegativeInteger |
- | Slow admission count. A non-negative Rust integer represented exactly as a JSON number. |
total_decode_forwards |
yes | NonNegativeInteger |
- | Total decode forwards. A non-negative Rust integer represented exactly as a JSON number. |
total_batched_decode_forwards |
yes | NonNegativeInteger |
- | Total batched decode forwards. A non-negative Rust integer represented exactly as a JSON number. |
total_decode_rows |
yes | NonNegativeInteger |
- | Total decode rows. A non-negative Rust integer represented exactly as a JSON number. |
total_prefill_admission_cycles |
yes | NonNegativeInteger |
- | Total prefill admission cycles. A non-negative Rust integer represented exactly as a JSON number. |
total_prefill_forwards |
yes | NonNegativeInteger |
- | Total prefill forwards. A non-negative Rust integer represented exactly as a JSON number. |
total_resident_prefill_attempts |
yes | NonNegativeInteger |
- | Total resident prefill attempts. A non-negative Rust integer represented exactly as a JSON number. |
total_resident_prefill_forwards |
yes | NonNegativeInteger |
- | Total resident prefill forwards. A non-negative Rust integer represented exactly as a JSON number. |
total_resident_prefill_initial_declines |
yes | NonNegativeInteger |
- | Total resident prefill initial declines. A non-negative Rust integer represented exactly as a JSON number. |
total_resident_prefill_route_failures |
yes | NonNegativeInteger |
- | Total resident prefill route failures. A non-negative Rust integer represented exactly as a JSON number. |
total_resident_prefill_rows |
yes | NonNegativeInteger |
- | Total resident prefill rows. A non-negative Rust integer represented exactly as a JSON number. |
total_resident_prefill_completed_rows |
yes | NonNegativeInteger |
- | Total resident prefill completed rows. A non-negative Rust integer represented exactly as a JSON number. |
last_resident_prefill_batch_size |
yes | NonNegativeInteger |
- | Last resident prefill batch size. A non-negative Rust integer represented exactly as a JSON number. |
max_resident_prefill_batch_size |
yes | NonNegativeInteger |
- | Max resident prefill batch size. A non-negative Rust integer represented exactly as a JSON number. |
total_decode_tokens |
yes | NonNegativeInteger |
- | Total decode tokens. A non-negative Rust integer represented exactly as a JSON number. |
total_prefill_tokens |
yes | NonNegativeInteger |
- | Total prefill tokens. A non-negative Rust integer represented exactly as a JSON number. |
total_prefill_layers |
yes | NonNegativeInteger |
- | Total prefill layers. A non-negative Rust integer represented exactly as a JSON number. |
total_prefill_layer_yields |
yes | NonNegativeInteger |
- | Total prefill layer yields. A non-negative Rust integer represented exactly as a JSON number. |
total_short_prefill_priority_forwards |
yes | NonNegativeInteger |
- | Total short prefill priority forwards. A non-negative Rust integer represented exactly as a JSON number. |
total_prefill_staging_priority_forwards |
yes | NonNegativeInteger |
- | Total prefill staging priority forwards. A non-negative Rust integer represented exactly as a JSON number. |
total_prefill_staging_admissions |
yes | NonNegativeInteger |
- | Total prefill staging admissions. A non-negative Rust integer represented exactly as a JSON number. |
total_errors |
yes | NonNegativeInteger |
- | Total errors. A non-negative Rust integer represented exactly as a JSON number. |
response_delivery_in_flight |
yes | NonNegativeInteger |
- | Response delivery in flight. A non-negative Rust integer represented exactly as a JSON number. |
response_delivery_backpressured |
yes | NonNegativeInteger |
- | Response delivery backpressured. A non-negative Rust integer represented exactly as a JSON number. |
response_delivery_pending_terminal |
yes | NonNegativeInteger |
- | Response delivery pending terminal. A non-negative Rust integer represented exactly as a JSON number. |
response_backpressure_events |
yes | NonNegativeInteger |
- | Response backpressure events. A non-negative Rust integer represented exactly as a JSON number. |
response_backpressure_wait_ms |
yes | NonNegativeInteger |
- | Response backpressure wait ms. A non-negative Rust integer represented exactly as a JSON number. |
response_stall_evictions |
yes | NonNegativeInteger |
- | Response stall evictions. A non-negative Rust integer represented exactly as a JSON number. |
response_channel_closed |
yes | NonNegativeInteger |
- | Response channel closed. A non-negative Rust integer represented exactly as a JSON number. |
adapter_groups_waiting |
yes | NonNegativeInteger |
- | Adapter groups waiting. A non-negative Rust integer represented exactly as a JSON number. |
prefix_deferred_waiting |
yes | NonNegativeInteger |
- | Prefix deferred waiting. A non-negative Rust integer represented exactly as a JSON number. |
prefix_admission_deferrals |
yes | NonNegativeInteger |
- | Prefix admission deferrals. A non-negative Rust integer represented exactly as a JSON number. |
last_forward_ms |
yes | NonNegativeNumber |
- | Last forward ms. A finite non-negative Rust floating-point value. |
max_decode_forward_ms |
yes | NonNegativeNumber |
- | Max decode forward ms. A finite non-negative Rust floating-point value. |
total_decode_forward_ms |
yes | NonNegativeNumber |
- | Total decode forward ms. A finite non-negative Rust floating-point value. |
last_prefill_ms |
yes | NonNegativeNumber |
- | Last prefill ms. A finite non-negative Rust floating-point value. |
max_prefill_forward_ms |
yes | NonNegativeNumber |
- | Max prefill forward ms. A finite non-negative Rust floating-point value. |
total_prefill_forward_ms |
yes | NonNegativeNumber |
- | Total prefill forward ms. A finite non-negative Rust floating-point value. |
last_admission_ms |
yes | NonNegativeNumber |
- | Last admission ms. A finite non-negative Rust floating-point value. |
max_admission_ms |
yes | NonNegativeNumber |
- | Max admission ms. A finite non-negative Rust floating-point value. |
total_admission_ms |
yes | NonNegativeNumber |
- | Total admission ms. A finite non-negative Rust floating-point value. |
total_actor_cycle_idle_ms |
yes | NonNegativeNumber |
- | Total actor cycle idle ms. A finite non-negative Rust floating-point value. |
max_actor_cycle_idle_ms |
yes | NonNegativeNumber |
- | Max actor cycle idle ms. A finite non-negative Rust floating-point value. |
stream_stall_grace_source |
yes | ConfigValueSource |
- | Stream stall grace source. The startup authority that supplied a resolved configuration value. |
actor_cycle_idle_source |
yes | ConfigValueSource |
- | Actor cycle idle source. The startup authority that supplied a resolved configuration value. |
actor_cycle_idle_active |
yes | Boolean |
- | Actor cycle idle active. A serialized Rust boolean. |
actor_barrier_adapter_active |
yes | Boolean |
- | Actor barrier adapter active. A serialized Rust boolean. |
actor_barrier_resize_active |
yes | Boolean |
- | Actor barrier resize active. A serialized Rust boolean. |
accepting |
yes | Boolean |
- | Accepting. A serialized Rust boolean. |
max_batch_tokens_source |
yes | ConfigValueSource |
- | Max batch tokens source. The startup authority that supplied a resolved configuration value. |
max_prefill_tokens_per_cycle_source |
yes | ConfigValueSource |
- | Max prefill tokens per cycle source. The startup authority that supplied a resolved configuration value. |
max_prefill_layers_per_cycle_source |
yes | ConfigValueSource |
- | Max prefill layers per cycle source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "BatchingEngineInfo | BatchingEngineSnapshotDebug"
}
BatchingRuntimeConfigRuntime-ready batching policy resolved once after backend selection.6 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
rowwise_decode |
yes | BatchingToggleDiagnostics |
- | Rowwise decode. A resolved batching toggle and its startup provenance. |
prefix_aware_admission |
yes | BatchingToggleDiagnostics |
- | Prefix aware admission. A resolved batching toggle and its startup provenance. |
prefill_admission_quantum |
yes | PrefillAdmissionQuantumDiagnostics |
- | Prefill admission quantum. Configured, backend, and effective prefill admission quantum. |
actor_cycle_idle |
yes | ActorCycleIdleDiagnostics |
- | Actor cycle idle. Configured cooperative safe-boundary actor idle and control-command polling contract. |
burst_prefill_admission |
yes | Boolean |
- | Burst prefill admission. A serialized Rust boolean. |
actor_prefill_tile_alignment_required |
yes | Boolean |
- | Actor prefill tile alignment required. A serialized Rust boolean. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "BatchingRuntimeConfig"
}
BatchingToggleDiagnosticsA resolved batching toggle and its startup provenance.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
enabled |
yes | Boolean |
- | Enabled. A serialized Rust boolean. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "BatchingToggleDiagnostics"
}
BooleanA serialized Rust boolean.0 fields
Type: boolean.
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "bool"
}
CacheDebugStateDeterministic response, prompt, token, prefix-cache, and backend recurrent-state ownership.9 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
deterministic_completion_entries |
yes | NonNegativeInteger |
- | Deterministic completion entries. A non-negative Rust integer represented exactly as a JSON number. |
deterministic_chat_request_entries |
yes | NonNegativeInteger |
- | Deterministic chat request entries. A non-negative Rust integer represented exactly as a JSON number. |
deterministic_chat_choices_entries |
yes | NonNegativeInteger |
- | Deterministic chat choices entries. A non-negative Rust integer represented exactly as a JSON number. |
deterministic_batch_entries |
yes | NonNegativeInteger |
- | Deterministic batch entries. A non-negative Rust integer represented exactly as a JSON number. |
batched_recurrent_state |
yes | BatchedStateCacheStats |
- | Batched recurrent state. Current ownership and process-lifetime lifecycle counters for the parked batched recurrent-state cache. |
resident_recurrent_state |
yes | GdnRecurrentStateResidencyStats |
- | Resident recurrent state. Current direct backend-private GDN recurrent-state ownership across resumable prefill and scoped decode rows; the persistent batched cache is separate. |
rendered_prompt |
yes | PromptCacheInfo |
- | Rendered prompt. One bounded prompt-cache hit and occupancy snapshot. |
prompt_token |
yes | PromptCacheInfo |
- | Prompt token. One bounded prompt-cache hit and occupancy snapshot. |
prefix_cache |
yes | PrefixCacheInfo |
- | Prefix cache. Prefix-cache occupancy, hit, limit, and lease counters. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "CacheDebugState"
}
CacheStatsPersistent logit-cache occupancy keyed by authoritative teacher revision.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
total_entries |
yes | NonNegativeInteger |
- | Total entries. A non-negative Rust integer represented exactly as a JSON number. |
total_bytes |
yes | NonNegativeInteger |
- | Total bytes. A non-negative Rust integer represented exactly as a JSON number. |
per_teacher |
yes | object |
- | Per teacher. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "CacheStats"
}
CacheStatsResponsePersistent logit-cache root and occupancy report.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
root |
yes | String |
- | Root. A serialized UTF-8 Rust string. |
stats |
yes | CacheStats |
- | Stats. Persistent logit-cache occupancy keyed by authoritative teacher revision. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "CacheStatsResponse"
}
CheckpointBoundaryPolicyResolved checkpoint-boundary retention and sparse replay policy.4 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
recompute_mode |
yes | CheckpointBoundaryRecomputeMode |
- | Recompute mode. Checkpoint-boundary retention or sparse replay mode. |
recompute_threshold_tokens |
yes | PositiveInteger |
- | Recompute threshold tokens. A positive Rust integer represented exactly as a JSON number. |
anchor_stride |
yes | PositiveInteger | null |
- | Anchor stride. |
cache_target_bytes |
yes | PositiveInteger |
- | Cache target bytes. A positive Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "CheckpointBoundaryPolicy"
}
CheckpointBoundaryRecomputeModeCheckpoint-boundary retention or sparse replay mode.0 fields
Type: string. Constraints: enum "auto", "enabled", "disabled".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "CheckpointBoundaryRecomputeMode"
}
CheckpointReadConfigResponseResolved startup-only checkpoint-read policy and completed real-model observations.10 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured_mib_per_second |
yes | PositiveInteger | null |
- | Configured MiB per second. |
rate_limited |
yes | Boolean |
- | Rate limited. A serialized Rust boolean. |
applicable |
yes | Boolean |
- | Applicable. A serialized Rust boolean. |
not_applicable_reason |
yes | any | null |
- | Not applicable reason. |
phases |
yes | any |
const ["snapshot_copy","initial_content_verification","post_upload_content_verification"] | Phases. |
cancellation_poll_milliseconds |
yes | any |
const 25 | Cancellation poll milliseconds. |
current_work_quantum_interruptible |
yes | any |
const false | Current work quantum interruptible. |
active_during_inference |
yes | any |
const false | Active during inference. |
restart_required_to_change |
yes | any |
const true | Restart required to change. |
observed |
yes | CheckpointReadReport | null |
- | Observed. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "CheckpointReadConfigResponse"
}
CheckpointReadPhaseReportCompleted logical/read-byte, elapsed-time, and pacing accounting for one checkpoint-read phase.7 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
stage |
yes | NonEmptyString |
- | Stage. A non-empty serialized UTF-8 Rust string. |
logical_bytes_completed |
yes | NonNegativeInteger |
- | Logical bytes completed. A non-negative Rust integer represented exactly as a JSON number. |
logical_bytes_total |
yes | NonNegativeInteger |
- | Logical bytes total. A non-negative Rust integer represented exactly as a JSON number. |
rate_limited_bytes_completed |
yes | NonNegativeInteger |
- | Rate limited bytes completed. A non-negative Rust integer represented exactly as a JSON number. |
elapsed_milliseconds |
yes | NonNegativeInteger |
- | Elapsed milliseconds. A non-negative Rust integer represented exactly as a JSON number. |
paced_milliseconds |
yes | NonNegativeInteger |
- | Paced milliseconds. A non-negative Rust integer represented exactly as a JSON number. |
complete |
yes | any |
const true | Complete. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "kiln_model::CheckpointReadPhaseReport"
}
CheckpointReadReportExact completed accounting for all loader-owned checkpoint-read phases.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured_bytes_per_second |
yes | PositiveInteger | null |
- | Configured bytes per second. |
snapshot_copy |
yes | CheckpointReadPhaseReport |
- | Snapshot copy. Completed logical/read-byte, elapsed-time, and pacing accounting for one checkpoint-read phase. |
initial_content_verification |
yes | CheckpointReadPhaseReport |
- | Initial content verification. Completed logical/read-byte, elapsed-time, and pacing accounting for one checkpoint-read phase. |
post_upload_content_verification |
yes | CheckpointReadPhaseReport |
- | Post upload content verification. Completed logical/read-byte, elapsed-time, and pacing accounting for one checkpoint-read phase. |
complete |
yes | any |
const true | Complete. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "kiln_model::CheckpointReadReport"
}
ConfigHashesAvailable hashes of loaded model, tokenizer, template, and effective environment configuration.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
tokenizer_config_hash |
no | Sha256 |
- | Tokenizer config hash. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
chat_template_hash |
no | Sha256 |
- | Chat template hash. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
training_chat_template_hash |
no | Sha256 |
- | Training chat template hash. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
model_config_hash |
no | Sha256 |
- | Model config hash. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
effective_config_hash |
no | Sha256 |
- | Effective config hash. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ConfigHashes"
}
ConfigResponseComplete resolved configuration and live capacity response.21 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
effective_configuration |
yes | EffectiveConfiguration |
- | Effective configuration. The deterministic flat map of every typed startup leaf after precedence, plus dynamic teacher-credential leaves. |
serving_profile |
yes | ServingProfileDiagnostics |
- | Serving profile. Operator-facing resolution report for process-lifetime serving policy. |
accelerator_runtime |
yes | ResolvedAcceleratorRuntimePolicy |
- | Accelerator runtime. Versioned process-lifetime accelerator policy. |
rocm_graphs |
yes | RocmGraphStats | null |
- | Point-in-time ROCm graph statistics, or null when a nonblocking snapshot is unavailable. |
rocm_graphs_unavailable_reason |
yes | RocmGraphUnavailableReason | null |
- | Closed reason rocm_graphs is null; null when statistics are present. |
rocm_graph_telemetry |
yes | RocmGraphLiveTelemetry | null |
- | Lock-independent ROCm graph phase telemetry, or null when no graph runner exists. |
rocm_graph_telemetry_unavailable_reason |
yes | RocmGraphUnavailableReason | null |
- | Closed reason rocm_graph_telemetry is null; null when telemetry is present. |
cuda_graphs |
yes | CudaGraphConfigResponse |
- | CUDA graphs. Startup-resolved CUDA graph request, cache bound, and fixed safety invariants. |
decode_runtime |
yes | DecodeRuntimeConfig |
- | Decode runtime. The immutable process-lifetime decode policy. |
batching |
yes | BatchingConfigResponse |
- | Batching. Resolved batching policy plus live actor state. |
prefix_cache |
yes | PrefixCacheConfigResponse |
- | Prefix cache. Requested prefix-cache policy plus backend-qualified live limits. |
streaming_prefill |
yes | StreamingPrefillRuntimeConfig |
- | Streaming prefill. Complete operator-facing resolved streaming-prefill policy. |
speculative |
yes | SpeculativeConfig |
- | Speculative. Configured, effective, and currently routable speculative-decoding state. |
model_startup |
yes | ModelStartupConfigResponse |
- | Model startup. Resolved model-startup resource policy and observations. |
paths |
yes | ResolvedApplicationPaths |
- | Paths. Absolute process-lifetime cache root and the startup authority that selected it. |
vram |
yes | VramConfig |
- | VRAM. Physical, configured, effective, and live VRAM capacity report. |
kv_cache |
yes | KvCacheConfig |
- | KV cache. Resolved KV-cache allocation and autoscaler state. |
training |
yes | TrainingConfigResponse |
- | Training. Resolved native-training capability and checkpoint configuration. |
memory_budget |
yes | MemoryBudgetConfig |
- | Memory budget. Startup model, KV-cache, training, and inference memory budget. |
generation |
yes | GenerationConfig |
- | Generation. Resolved server-wide thinking and reasoning-content defaults. |
operational |
yes | OperationalRuntimeConfig |
- | Operational. Startup-resolved operational paths and local-capability access policy retained by request handlers. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ConfigResponse"
}
ConfigValueSourceThe startup authority that supplied a resolved configuration value.0 fields
Type: string. Constraints: enum "default", "config_file", "environment", "command_line".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ConfigValueSource"
}
CudaFlashBackwardModeThe process-lifetime CUDA FlashAttention backward accumulation mode.0 fields
Type: string. Constraints: enum "fast", "deterministic".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "CudaFlashBackwardMode"
}
CudaGraphConfigResponseStartup-resolved CUDA graph request, cache bound, and fixed safety invariants.7 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
requested |
yes | Boolean |
- | Requested. A serialized Rust boolean. |
capture_allowed_by_serving_profile |
yes | Boolean |
- | Capture allowed by serving profile. A serialized Rust boolean. |
effective_policy_enabled |
yes | Boolean |
- | Effective policy enabled. A serialized Rust boolean. |
max_cached_graphs |
yes | PositiveInteger |
- | Max cached graphs. A positive Rust integer represented exactly as a JSON number. |
stable_paged_metadata |
yes | any |
const true | Stable paged metadata. |
batched_capture_available |
yes | any |
const false | Batched capture available. |
restart_required_to_change |
yes | any |
const true | Restart required to change. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "CudaGraphConfigResponse"
}
CudaGraphInfoConfigured CUDA graph request, live runner state, bounded cache, and fixed safety invariants.8 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
requested |
yes | Boolean |
- | Requested. A serialized Rust boolean. |
capture_allowed_by_serving_profile |
yes | Boolean |
- | Capture allowed by serving profile. A serialized Rust boolean. |
enabled |
yes | Boolean | null |
- | Whether CUDA graph replay is enabled; null while the nonblocking snapshot is busy. |
state |
yes | string |
enum "enabled", "disabled", "busy" | State. |
max_cached_graphs |
yes | PositiveInteger |
- | Max cached graphs. A positive Rust integer represented exactly as a JSON number. |
stable_paged_metadata |
yes | any |
const true | Stable paged metadata. |
batched_capture_available |
yes | any |
const false | Batched capture available. |
restart_required_to_change |
yes | any |
const true | Restart required to change. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "CudaGraphInfo"
}
CudaKernelProfileThe process-lifetime CUDA backend-kernel route set.0 fields
Type: string. Constraints: enum "native_default", "portable_fallback".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "CudaKernelProfile"
}
CudaMarlinProfileThe process-lifetime CUDA Marlin projection layout.0 fields
Type: string. Constraints: enum "disabled", "attention_mlp", "attention_mlp_gdn".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "CudaMarlinProfile"
}
CudaSynchronizationReasonClosed reason vocabulary for CUDA host synchronization.0 fields
Type: string. Constraints: enum "explicit_device_drain", "explicit_stream_drain", "tensor_handoff", "external_yield", "in_place_mutation", "memory_reclaim", "graph_boundary", "full_attention_handoff", "model_handoff", "host_readback", "allocation_lifetime", "global_state_mutation".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "CudaSyncReason"
}
CudaSynchronizationReasonStatsFixed-cardinality counters for one CUDA synchronization reason.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
reason |
yes | CudaSynchronizationReason |
- | Reason. Closed reason vocabulary for CUDA host synchronization. |
device_wait_count |
yes | NonNegativeInteger |
- | Device wait count. A non-negative Rust integer represented exactly as a JSON number. |
stream_wait_count |
yes | NonNegativeInteger |
- | Stream wait count. A non-negative Rust integer represented exactly as a JSON number. |
failure_count |
yes | NonNegativeInteger |
- | Failure count. A non-negative Rust integer represented exactly as a JSON number. |
waited_ns |
yes | NonNegativeInteger |
- | Waited ns. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "CudaSynchronizationReasonStats"
}
CudaSynchronizationRuntimeStatsPoint-in-time CUDA synchronization telemetry.8 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
active |
yes | Boolean |
- | Active. A serialized Rust boolean. |
telemetry_available |
yes | Boolean |
- | Telemetry available. A serialized Rust boolean. |
telemetry_error |
no | String |
- | Telemetry error. A serialized UTF-8 Rust string. |
total_device_wait_count |
yes | NonNegativeInteger |
- | Total device wait count. A non-negative Rust integer represented exactly as a JSON number. |
total_stream_wait_count |
yes | NonNegativeInteger |
- | Total stream wait count. A non-negative Rust integer represented exactly as a JSON number. |
total_failure_count |
yes | NonNegativeInteger |
- | Total failure count. A non-negative Rust integer represented exactly as a JSON number. |
total_waited_ns |
yes | NonNegativeInteger |
- | Total waited ns. A non-negative Rust integer represented exactly as a JSON number. |
reasons |
yes | array<CudaSynchronizationReasonStats> | array<CudaSynchronizationReasonStats> |
- | Reasons. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "CudaSynchronizationRuntimeStats"
}
DebugDisabledResponseHTTP 403 body returned when model-state diagnostics are not enabled.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
error |
yes | any |
const "debug endpoint disabled" | Stable error code for a disabled debug endpoint. |
enable_with |
yes | any |
const "set server.debug_model_state=true or server.eval_mode=true" | Configuration change required before restart to enable the endpoint. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "DebugDisabledResponse"
}
DebugProvenanceErrorResponseHTTP 500 body returned when the resident execution provenance fails validation.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
error |
yes | any |
const "invalid execution provenance" | Stable error code for resident provenance that failed validation. |
detail |
yes | NonEmptyString |
- | Detail. A non-empty serialized UTF-8 Rust string. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "serde_json::Value"
}
DecodeBatchEffectiveSourceThe authority that selected the effective concurrent decode width.0 fields
Type: string. Constraints: enum "backend_policy", "config_file", "environment", "command_line", "deterministic", "max_batch_tokens".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "DecodeBatchEffectiveSource"
}
DecodeRuntimeConfigThe immutable process-lifetime decode policy.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
deterministic |
yes | DeterministicInferenceDiagnostics |
- | Deterministic. Resolved deterministic-inference state and startup provenance. |
max_decode_batch |
yes | MaxDecodeBatchDiagnostics |
- | Max decode batch. Configured, backend, and effective concurrent decode width. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "DecodeRuntimeConfig"
}
DecodeRuntimeInfoResolved decode configuration and live backend execution state.11 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configuration |
yes | DecodeRuntimeConfig |
- | Configuration. The immutable process-lifetime decode policy. |
accelerator_runtime |
yes | ResolvedAcceleratorRuntimePolicy |
- | Accelerator runtime. Versioned process-lifetime accelerator policy. |
cuda_synchronization |
yes | CudaSynchronizationRuntimeStats |
- | CUDA synchronization. Point-in-time CUDA synchronization telemetry. |
rocm_synchronization |
yes | RocmSynchronizationRuntimeStats |
- | ROCm synchronization. Point-in-time ROCm synchronization telemetry. |
batching_configuration |
yes | BatchingRuntimeConfig |
- | Batching configuration. Runtime-ready batching policy resolved once after backend selection. |
cuda_graphs |
yes | CudaGraphInfo |
- | CUDA graphs. Configured CUDA graph request, live runner state, bounded cache, and fixed safety invariants. |
rocm_graphs |
yes | RocmGraphInfo |
- | ROCm graphs. Nonblocking ROCm graph state projected for health reporting. |
metal_graphs |
yes | GraphInfo |
- | Metal graphs. Nonblocking CUDA or Metal graph state. |
kv_autoscaler |
yes | KvAutoscalerState |
- | KV autoscaler. Startup and activation state for the physical KV-cache autoscaler. |
memory_governor |
yes | MemoryGovernorRuntimeInfo |
- | Memory governor. Live memory-governor mode, sampler health, and automatic reclaim counters. |
batching_engine |
yes | BatchingEngineSnapshot | null |
- | Batching engine. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "DecodeRuntimeInfo"
}
DecodeStatsSnapshotRolling request-local decode throughput, ITL tail, and bounded stall diagnosis.12 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
tok_per_sec |
yes | NonNegativeNumber |
- | Tok per sec. A finite non-negative Rust floating-point value. |
p50_itl_ms |
yes | NonNegativeNumber |
- | P50 ITL ms. A finite non-negative Rust floating-point value. |
p99_itl_ms |
yes | NonNegativeNumber |
- | P99 ITL ms. A finite non-negative Rust floating-point value. |
p999_itl_ms |
yes | NonNegativeNumber |
- | P999 ITL ms. A finite non-negative Rust floating-point value. |
mean_itl_ms |
yes | NonNegativeNumber |
- | Mean ITL ms. A finite non-negative Rust floating-point value. |
max_itl_ms |
yes | NonNegativeNumber |
- | Max ITL ms. A finite non-negative Rust floating-point value. |
stall_threshold_ms |
yes | NonNegativeNumber |
- | Stall threshold ms. A finite non-negative Rust floating-point value. |
stall_count |
yes | NonNegativeInteger |
- | Stall count. A non-negative Rust integer represented exactly as a JSON number. |
unexplained_stall_count |
yes | NonNegativeInteger |
- | Unexplained stall count. A non-negative Rust integer represented exactly as a JSON number. |
stall_reasons |
yes | LatencyStallReasonCounts |
- | Stall reasons. Fixed-cardinality token-stall counts by dominant blocking reason. |
sample_count |
yes | NonNegativeInteger |
- | Sample count. A non-negative Rust integer represented exactly as a JSON number. |
window_secs |
yes | any |
const 60 | Fixed rolling window in seconds. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "DecodeStatsSnapshot"
}
DeterministicInferenceDiagnosticsResolved deterministic-inference state and startup provenance.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
enabled |
yes | Boolean |
- | Enabled. A serialized Rust boolean. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "DeterministicInferenceDiagnostics"
}
EffectiveConfigurationThe deterministic flat map of every typed startup leaf after precedence, plus dynamic teacher-credential leaves.7 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
schema_id |
yes | any |
const "kiln.effective-configuration.v1" | Schema ID. |
schema_version |
yes | any |
const 1 | Schema version. |
effective_config_hash |
yes | Sha256 | null |
- | Effective config hash. |
fixed_field_count |
yes | any |
const 118 | Fixed field count. |
dynamic_field_count |
yes | NonNegativeInteger |
- | Dynamic field count. A non-negative Rust integer represented exactly as a JSON number. |
all_fields_restart_required_to_change |
yes | any |
const true | All fields restart required to change. |
fields |
yes | object |
- | Fields. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "config::EffectiveConfiguration"
}
EffectiveConfigurationFieldOne post-precedence typed startup value, its authority, environment spellings, redaction state, and lifecycle.6 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
effective_value |
yes | any |
- | Effective value. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
canonical_environment |
yes | String | null |
- | Canonical environment. |
compatibility_environment |
yes | array<String> |
- | Compatibility environment. |
redacted |
yes | Boolean |
- | Redacted. A serialized Rust boolean. |
restart_required_to_change |
yes | any |
const true | Restart required to change. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "config::EffectiveConfigurationField"
}
ExecutionBackendIdentityBackend and numerical-runtime identity.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
name |
yes | NonEmptyString |
- | Name. A non-empty serialized UTF-8 Rust string. |
device |
yes | NonEmptyString |
- | Device. A non-empty serialized UTF-8 Rust string. |
numerical_runtime_sha256 |
yes | Sha256 |
- | Numerical runtime SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ExecutionBackendIdentity"
}
ExecutionBuildIdentityExecutable build identity.6 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
package_version |
yes | NonEmptyString |
- | Package version. A non-empty serialized UTF-8 Rust string. |
target |
yes | NonEmptyString |
- | Target. A non-empty serialized UTF-8 Rust string. |
executable_sha256 |
yes | Sha256 |
- | Executable SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
git_commit |
no | NonEmptyString |
- | Git commit. A non-empty serialized UTF-8 Rust string. |
source_tree_sha256 |
no | Sha256 |
- | Source tree SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
source_dirty |
no | Boolean |
- | Source dirty. A serialized Rust boolean. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ExecutionBuildIdentity"
}
ExecutionConfigurationIdentityEffective server configuration and environment identity.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
effective_server_config_sha256 |
yes | Sha256 |
- | Effective server config SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
effective_environment_sha256 |
yes | Sha256 |
- | Effective environment SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ExecutionConfigurationIdentity"
}
ExecutionIdentitySummaryCompact execution-provenance identity.11 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
provenance_type |
yes | String |
- | Provenance type. A serialized UTF-8 Rust string. |
backend |
yes | String |
- | Backend. A serialized UTF-8 Rust string. |
device |
yes | String |
- | Device. A serialized UTF-8 Rust string. |
inference_dtype |
yes | String |
- | Inference dtype. A serialized UTF-8 Rust string. |
training_policy |
yes | String |
- | Training policy. A serialized UTF-8 Rust string. |
provenance_sha256 |
yes | Sha256 |
- | Provenance SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
executable_sha256 |
yes | Sha256 |
- | Executable SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
numerical_runtime_sha256 |
yes | Sha256 |
- | Numerical runtime SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
kernel_contract_sha256 |
yes | Sha256 |
- | Kernel contract SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
effective_server_config_sha256 |
yes | Sha256 |
- | Effective server config SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
effective_environment_sha256 |
yes | Sha256 |
- | Effective environment SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ExecutionIdentitySummary"
}
ExecutionKernelIdentityVersioned compiled-kernel contract identity.4 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
contract_type |
yes | any |
const "kiln.kernel-contract.v1" | Contract type. |
versions |
yes | object |
- | Versions. |
compiled_features |
yes | array<NonEmptyString> |
- | Compiled features. |
contract_sha256 |
yes | Sha256 |
- | Contract SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ExecutionKernelIdentity"
}
ExecutionModelIdentityLoaded model, tokenizer, and template identity.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
model_config_sha256 |
yes | Sha256 |
- | Model config SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
tokenizer_vocab_sha256 |
yes | Sha256 |
- | Tokenizer vocab SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
tokenizer_config_sha256 |
yes | Sha256 |
- | Tokenizer config SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
chat_template_sha256 |
no | Sha256 |
- | Chat template SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
training_chat_template_sha256 |
no | Sha256 |
- | Training chat template SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ExecutionModelIdentity"
}
ExecutionPrecisionIdentityResolved inference and training precision identity.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
inference_dtype |
yes | NonEmptyString |
- | Inference dtype. A non-empty serialized UTF-8 Rust string. |
training_policy |
yes | NonEmptyString |
- | Training policy. A non-empty serialized UTF-8 Rust string. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ExecutionPrecisionIdentity"
}
ExecutionProvenanceV1Immutable self-verifying execution envelope captured at server startup.9 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
schema_version |
yes | any |
const 1 | Schema version. |
provenance_type |
yes | any |
const "kiln.execution-provenance.v1" | Provenance type. |
backend |
yes | ExecutionBackendIdentity |
- | Backend. Backend and numerical-runtime identity. |
build |
yes | ExecutionBuildIdentity |
- | Build. Executable build identity. |
model |
yes | ExecutionModelIdentity |
- | Model. Loaded model, tokenizer, and template identity. |
precision |
yes | ExecutionPrecisionIdentity |
- | Precision. Resolved inference and training precision identity. |
kernels |
yes | ExecutionKernelIdentity |
- | Kernels. Versioned compiled-kernel contract identity. |
configuration |
yes | ExecutionConfigurationIdentity |
- | Configuration. Effective server configuration and environment identity. |
provenance_sha256 |
yes | Sha256 |
- | Provenance SHA-256. A lowercase SHA-256 digest with Kiln's sha256: prefix. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ExecutionProvenanceV1"
}
ExternalYieldSyncStatsProcess-lifetime settlement counters at one external-yield boundary.6 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
boundary |
yes | String |
- | Boundary. A serialized UTF-8 Rust string. |
calls |
yes | NonNegativeInteger |
- | Calls. A non-negative Rust integer represented exactly as a JSON number. |
failures |
yes | NonNegativeInteger |
- | Failures. A non-negative Rust integer represented exactly as a JSON number. |
total_micros |
yes | NonNegativeInteger |
- | Total micros. A non-negative Rust integer represented exactly as a JSON number. |
max_micros |
yes | NonNegativeInteger |
- | Max micros. A non-negative Rust integer represented exactly as a JSON number. |
slow_calls |
yes | NonNegativeInteger |
- | Slow calls. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ExternalYieldSyncStats"
}
FiniteNumberA finite Rust floating-point value. serde_json rejects non-finite values.0 fields
Type: number.
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "f32 | f64"
}
GdnRecurrentStateResidencyStatsCurrent direct backend-private GDN recurrent-state ownership across resumable prefill and scoped decode rows; the persistent batched cache is separate.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
entry_count |
yes | NonNegativeInteger |
- | Entry count. A non-negative Rust integer represented exactly as a JSON number. |
buffer_bytes |
yes | NonNegativeInteger |
- | Buffer bytes. A non-negative Rust integer represented exactly as a JSON number. |
allocation_bytes |
yes | NonNegativeInteger |
- | Allocation bytes. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "GdnRecurrentStateResidencyStats"
}
GenerationConfigResolved server-wide thinking and reasoning-content defaults.4 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
default_thinking_enabled |
yes | Boolean | null |
- | Server-wide thinking default; null preserves the model template's default. |
default_thinking_budget_tokens |
yes | NonNegativeInteger | null |
- | Default thinking-token limit; null means unlimited. |
default_thinking_budget_ms |
yes | NonNegativeInteger | null |
- | Default thinking-time limit in milliseconds; null means unlimited. |
fold_reasoning_into_content |
yes | Boolean |
- | Fold reasoning into content. A serialized Rust boolean. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "GenerationConfig"
}
GpuMemoryInfoStartup GPU memory budget plus the live all-process memory observation.20 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
total_vram_bytes |
yes | NonNegativeInteger |
- | Total VRAM bytes. A non-negative Rust integer represented exactly as a JSON number. |
model_bytes |
yes | NonNegativeInteger |
- | Model bytes. A non-negative Rust integer represented exactly as a JSON number. |
estimated_model_bytes |
yes | NonNegativeInteger |
- | Estimated model bytes. A non-negative Rust integer represented exactly as a JSON number. |
post_load_used_bytes |
yes | NonNegativeInteger |
- | Post load used bytes. A non-negative Rust integer represented exactly as a JSON number. |
peak_prefill_used_bytes |
yes | NonNegativeInteger |
- | Peak prefill used bytes. A non-negative Rust integer represented exactly as a JSON number. |
kv_cache_bytes |
yes | NonNegativeInteger |
- | KV cache bytes. A non-negative Rust integer represented exactly as a JSON number. |
training_budget_bytes |
yes | NonNegativeInteger |
- | Training budget bytes. A non-negative Rust integer represented exactly as a JSON number. |
allocated_bytes |
yes | NonNegativeInteger |
- | Allocated bytes. A non-negative Rust integer represented exactly as a JSON number. |
reserved_bytes |
yes | NonNegativeInteger |
- | Reserved bytes. A non-negative Rust integer represented exactly as a JSON number. |
total_vram_gb |
yes | NonNegativeNumber |
- | Total VRAM gb. A finite non-negative Rust floating-point value. |
model_gb |
yes | NonNegativeNumber |
- | Model gb. A finite non-negative Rust floating-point value. |
estimated_model_gb |
yes | NonNegativeNumber |
- | Estimated model gb. A finite non-negative Rust floating-point value. |
post_load_used_gb |
yes | NonNegativeNumber |
- | Post load used gb. A finite non-negative Rust floating-point value. |
peak_prefill_used_gb |
yes | NonNegativeNumber |
- | Peak prefill used gb. A finite non-negative Rust floating-point value. |
kv_cache_gb |
yes | NonNegativeNumber |
- | KV cache gb. A finite non-negative Rust floating-point value. |
training_budget_gb |
yes | NonNegativeNumber |
- | Training budget gb. A finite non-negative Rust floating-point value. |
allocated_gb |
yes | NonNegativeNumber |
- | Allocated gb. A finite non-negative Rust floating-point value. |
reserved_gb |
yes | NonNegativeNumber |
- | Reserved gb. A finite non-negative Rust floating-point value. |
inference_memory_fraction |
yes | NonNegativeNumber |
- | Inference memory fraction. A finite non-negative Rust floating-point value. |
live |
yes | LiveMemory | null |
- | Live. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "GpuMemoryInfo"
}
GradientCheckpointPolicyTagged immutable gradient-checkpoint behavior for one native training run.0 fields
Type: object | object | object.
Composition and conditional rules
The following JSON is copied exactly from this schema node.
{
"oneOf": [
{
"type": "object",
"additionalProperties": false,
"required": [
"mode"
],
"properties": {
"mode": {
"const": "auto"
}
}
},
{
"type": "object",
"additionalProperties": false,
"required": [
"mode",
"segments"
],
"properties": {
"mode": {
"const": "explicit_segments"
},
"segments": {
"$ref": "#/$defs/PositiveInteger"
}
}
},
{
"type": "object",
"additionalProperties": false,
"required": [
"mode",
"segments"
],
"properties": {
"mode": {
"const": "disabled"
},
"segments": {
"anyOf": [
{
"$ref": "#/$defs/PositiveInteger"
},
{
"type": "null"
}
]
}
}
}
]
}
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "GradientCheckpointPolicy"
}
GraphInfoNonblocking CUDA or Metal graph state.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
enabled |
yes | Boolean | null |
- | Enabled. |
state |
yes | string |
enum "enabled", "disabled", "busy" | State. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "GraphInfo"
}
HealthCheckOne stable readiness condition and its result.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
name |
yes | string |
enum "model_loaded", "scheduler_responsive", "backend_runtime_healthy", "inference_admission", "inference_prewarm_complete", "execution_provenance_valid" | Name. |
pass |
yes | Boolean |
- | Pass. A serialized Rust boolean. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "HealthCheck"
}
HealthResponseComplete readiness and runtime diagnostic response returned with HTTP 200 or 503.37 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
status |
yes | string |
enum "ok", "degraded", "maintenance" | ok is ready for normal work, degraded reports a failed readiness check, and maintenance reports an intentionally drained server. |
version |
yes | NonEmptyString |
- | Version. A non-empty serialized UTF-8 Rust string. |
uptime_seconds |
yes | NonNegativeInteger |
- | Uptime seconds. A non-negative Rust integer represented exactly as a JSON number. |
model |
yes | String |
- | Model. A serialized UTF-8 Rust string. |
backend |
yes | string |
enum "mock", "model" | Whether the server is using mock responses or a loaded model. |
backend_runtime |
yes | BackendRuntimeInfo |
- | Backend runtime. Backend health, quarantine, restart, and external-yield settlement state. |
serving_profile |
yes | ServingProfileDiagnostics |
- | Serving profile. Operator-facing resolution report for process-lifetime serving policy. |
http |
yes | HttpRuntimeInfo |
- | HTTP. Resolved and kernel-readback HTTP socket send-buffer state. |
model_defaults_profile |
yes | ModelDefaultsProfile |
- | Model defaults profile. Built-in runtime defaults for the supported model family. |
eval_mode |
yes | Boolean |
- | Eval mode. A serialized Rust boolean. |
debug_model_state |
yes | Boolean |
- | Debug model state. A serialized Rust boolean. |
default_thinking_enabled |
yes | Boolean | null |
- | Server-wide thinking default; null preserves the model template's default. |
default_thinking_budget_tokens |
yes | NonNegativeInteger | null |
- | Default thinking-token limit; null means unlimited. |
default_thinking_budget_ms |
yes | NonNegativeInteger | null |
- | Default thinking-time limit in milliseconds; null means unlimited. |
fold_reasoning_into_content |
yes | Boolean |
- | Fold reasoning into content. A serialized Rust boolean. |
config_hashes |
yes | ConfigHashes |
- | Config hashes. Available hashes of loaded model, tokenizer, template, and effective environment configuration. |
base_weight_identity |
yes | BaseWeightIdentitySummary | null |
- | Loaded base-weight identity; null in mock mode. |
execution_identity |
yes | ExecutionIdentitySummary | null |
- | Validated execution identity; null when no real-model provenance is resident. |
active_adapter |
yes | String | null |
- | Adapter selected for new inference requests; null selects the base model. |
loaded_adapter |
yes | String | null |
- | Adapter currently resident in the model runner; null when only base weights are loaded. |
loaded_adapter_revision |
yes | String | null |
- | Content revision of the resident adapter; null when no adapter is loaded. |
loaded_adapter_count |
yes | NonNegativeInteger |
- | Loaded adapter count. A non-negative Rust integer represented exactly as a JSON number. |
adapters_loaded |
yes | NonNegativeInteger |
- | Adapters loaded. A non-negative Rust integer represented exactly as a JSON number. |
request_count |
yes | NonNegativeInteger |
- | Request count. A non-negative Rust integer represented exactly as a JSON number. |
requests |
yes | RequestMetrics |
- | Requests. Lifetime request outcome and concurrency counters. |
recent_requests |
yes | RecentRequestMetrics |
- | Recent requests. Aggregated metrics over the bounded recent-request ring. |
scheduler |
yes | SchedulerStats | null |
- | Paged-KV scheduler gauges; null when the active backend does not expose them. |
self_improve_scheduler |
no | SelfImproveSchedulerStatus |
- | Self improve scheduler. Durable self-improvement scheduler status. |
gpu_memory |
yes | GpuMemoryInfo | null |
- | GPU capacity and live all-process memory observation; null without a GPU backend. |
vulkan_buffers |
no | VulkanBufferInfo |
- | Vulkan buffers. Process-lifetime live and cumulative VulkanBuffer allocation accounting by memory route. |
vulkan_buffer_pool |
no | VulkanBufferPoolInfo |
- | Vulkan buffer pool. Bounded Vulkan scratch-recycler ownership, effectiveness, eviction, and overflow accounting. |
prefix_cache |
yes | PrefixCacheHealthInfo |
- | Prefix cache. Prefix-cache counters plus normalized utilization for health reporting. |
prompt_caches |
yes | PromptCachesInfo |
- | Prompt caches. Rendered-prompt and prompt-token cache snapshots. |
decode_runtime |
yes | DecodeRuntimeInfo |
- | Decode runtime. Resolved decode configuration and live backend execution state. |
prefill_runtime |
yes | PrefillRuntimeInfo |
- | Prefill runtime. Resolved prefill execution state. |
training |
yes | TrainingInfo |
- | Training. Live training queue and checkpoint-boundary state. |
checks |
yes | array<HealthCheck> |
minimum items 6; maximum items 6 | The six stable readiness checks whose failures determine degraded status and HTTP 503. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "HealthResponse"
}
HttpRuntimeInfoResolved and kernel-readback HTTP socket send-buffer state.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
send_buffer_requested_bytes |
yes | NonNegativeInteger | null |
- | Configured send-buffer target in bytes, or null when Kiln leaves the socket default unchanged. |
send_buffer_kernel_readback_bytes |
yes | NonNegativeInteger | null |
- | Send-buffer size reported by the operating system after listener setup, or null when no target was requested. |
send_buffer_effective_bytes |
yes | NonNegativeInteger | null |
- | Effective send-buffer target used for startup validation, or null when Kiln leaves the socket default unchanged. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "HttpRuntimeInfo | HttpDebugState"
}
KtApiModeThe process-lifetime kiln-tensor adapter route selection.0 fields
Type: string. Constraints: enum "auto", "all", "disabled".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "KtApiMode"
}
KvAutoscalerStateStartup and activation state for the physical KV-cache autoscaler.10 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
requested |
yes | Boolean |
- | Requested. A serialized Rust boolean. |
requested_source |
yes | ConfigValueSource |
- | Requested source. The startup authority that supplied a resolved configuration value. |
force_blocks |
yes | NonNegativeInteger | null |
- | Force blocks. |
force_blocks_source |
yes | ConfigValueSource |
- | Force blocks source. The startup authority that supplied a resolved configuration value. |
enabled |
yes | Boolean |
- | Enabled. A serialized Rust boolean. |
state |
yes | String |
- | State. A serialized UTF-8 Rust string. |
reason |
yes | String |
- | Reason. A serialized UTF-8 Rust string. |
start_blocks |
yes | NonNegativeInteger | null |
- | Start blocks. |
min_blocks |
yes | NonNegativeInteger | null |
- | Min blocks. |
bytes_per_block |
yes | NonNegativeInteger | null |
- | Bytes per block. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "KvAutoscalerState"
}
KvCacheConfigResolved KV-cache allocation and autoscaler state.4 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
num_blocks |
yes | NonNegativeInteger |
- | Num blocks. A non-negative Rust integer represented exactly as a JSON number. |
num_blocks_source |
yes | String |
- | Num blocks source. A serialized UTF-8 Rust string. |
fp8_enabled |
yes | Boolean |
- | FP8 enabled. A serialized Rust boolean. |
autoscaler |
yes | KvAutoscalerState |
- | Autoscaler. Startup and activation state for the physical KV-cache autoscaler. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "KvCacheConfig"
}
LastErrorSummaryCompact summary of the most recent failed or timed-out request.6 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
id |
yes | String |
- | ID. A serialized UTF-8 Rust string. |
timestamp_unix_ms |
yes | NonNegativeInteger |
- | Timestamp unix ms. A non-negative Rust integer represented exactly as a JSON number. |
finish_reason |
yes | String |
- | Finish reason. A serialized UTF-8 Rust string. |
error |
yes | String | null |
- | Error. |
duration_ms |
yes | NonNegativeInteger |
- | Duration ms. A non-negative Rust integer represented exactly as a JSON number. |
adapter |
yes | String | null |
- | Adapter. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "LastErrorSummary"
}
LatencyPercentilesLatency percentiles in milliseconds.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
p50 |
yes | NonNegativeNumber |
- | P50. A finite non-negative Rust floating-point value. |
p95 |
yes | NonNegativeNumber |
- | P95. A finite non-negative Rust floating-point value. |
p99 |
yes | NonNegativeNumber |
- | P99. A finite non-negative Rust floating-point value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "LatencyPercentiles"
}
LatencyPhaseTimingsBounded request phase timings; null means the serving path did not measure that subphase.20 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
actor_queue_ms |
yes | NonNegativeNumber | null |
- | Actor queue ms. |
actor_admission_ms |
yes | NonNegativeNumber | null |
- | Actor admission ms. |
tokenization_ms |
yes | NonNegativeNumber | null |
- | Tokenization ms. |
prefill_ms |
yes | NonNegativeNumber | null |
- | Prefill ms. |
decode_ms |
yes | NonNegativeNumber | null |
- | Decode ms. |
actor_cycle_idle_ms |
yes | NonNegativeNumber | null |
- | Actor cycle idle ms. |
sampling_ms |
yes | NonNegativeNumber | null |
- | Sampling ms. |
readback_ms |
yes | NonNegativeNumber | null |
- | Readback ms. |
response_delivery_ms |
yes | NonNegativeNumber | null |
- | Response delivery ms. |
handler_queue_ms |
yes | NonNegativeNumber | null |
- | Handler queue ms. |
client_delivery_ms |
yes | NonNegativeNumber | null |
- | Client delivery ms. |
gpu_lock_wait_ms |
yes | NonNegativeNumber | null |
- | GPU lock wait ms. |
graph_capture_ms |
yes | NonNegativeNumber | null |
- | Graph capture ms. |
graph_replay_ms |
yes | NonNegativeNumber | null |
- | Graph replay ms. |
synchronization_ms |
yes | NonNegativeNumber | null |
- | Synchronization ms. |
resize_ms |
yes | NonNegativeNumber | null |
- | Resize ms. |
trim_ms |
yes | NonNegativeNumber | null |
- | Trim ms. |
adapter_ms |
yes | NonNegativeNumber | null |
- | Adapter ms. |
training_ms |
yes | NonNegativeNumber | null |
- | Training ms. |
unexplained_ms |
yes | NonNegativeNumber | null |
- | Unexplained ms. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "LatencyPhaseTimings"
}
LatencyStallReasonCountsFixed-cardinality token-stall counts by dominant blocking reason.19 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
actor_queue |
yes | NonNegativeInteger |
- | Actor queue. A non-negative Rust integer represented exactly as a JSON number. |
actor_admission |
yes | NonNegativeInteger |
- | Actor admission. A non-negative Rust integer represented exactly as a JSON number. |
actor_prefill |
yes | NonNegativeInteger |
- | Actor prefill. A non-negative Rust integer represented exactly as a JSON number. |
actor_decode |
yes | NonNegativeInteger |
- | Actor decode. A non-negative Rust integer represented exactly as a JSON number. |
actor_cycle_idle |
yes | NonNegativeInteger |
- | Actor cycle idle. A non-negative Rust integer represented exactly as a JSON number. |
response_delivery |
yes | NonNegativeInteger |
- | Response delivery. A non-negative Rust integer represented exactly as a JSON number. |
handler_queue |
yes | NonNegativeInteger |
- | Handler queue. A non-negative Rust integer represented exactly as a JSON number. |
client_delivery |
yes | NonNegativeInteger |
- | Client delivery. A non-negative Rust integer represented exactly as a JSON number. |
sampling |
yes | NonNegativeInteger |
- | Sampling. A non-negative Rust integer represented exactly as a JSON number. |
readback |
yes | NonNegativeInteger |
- | Readback. A non-negative Rust integer represented exactly as a JSON number. |
gpu_lock_wait |
yes | NonNegativeInteger |
- | GPU lock wait. A non-negative Rust integer represented exactly as a JSON number. |
graph_capture |
yes | NonNegativeInteger |
- | Graph capture. A non-negative Rust integer represented exactly as a JSON number. |
graph_replay |
yes | NonNegativeInteger |
- | Graph replay. A non-negative Rust integer represented exactly as a JSON number. |
synchronization |
yes | NonNegativeInteger |
- | Synchronization. A non-negative Rust integer represented exactly as a JSON number. |
resize |
yes | NonNegativeInteger |
- | Resize. A non-negative Rust integer represented exactly as a JSON number. |
trim |
yes | NonNegativeInteger |
- | Trim. A non-negative Rust integer represented exactly as a JSON number. |
adapter |
yes | NonNegativeInteger |
- | Adapter. A non-negative Rust integer represented exactly as a JSON number. |
training |
yes | NonNegativeInteger |
- | Training. A non-negative Rust integer represented exactly as a JSON number. |
unexplained |
yes | NonNegativeInteger |
- | Unexplained. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "LatencyStallReasonCounts"
}
LiveMemoryThe memory governor's current all-process memory observation.16 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
probe_failed |
yes | Boolean |
- | Probe failed. A serialized Rust boolean. |
total_gb |
yes | NonNegativeNumber |
- | Total gb. A finite non-negative Rust floating-point value. |
used_gb |
yes | NonNegativeNumber |
- | Used gb. A finite non-negative Rust floating-point value. |
free_gb |
yes | NonNegativeNumber |
- | Free gb. A finite non-negative Rust floating-point value. |
available_gb |
yes | NonNegativeNumber |
- | Available gb. A finite non-negative Rust floating-point value. |
soft_reserved_gb |
yes | NonNegativeNumber |
- | Soft reserved gb. A finite non-negative Rust floating-point value. |
pressure |
yes | String |
- | Pressure. A serialized UTF-8 Rust string. |
source |
yes | String |
- | Source. A serialized UTF-8 Rust string. |
unified |
yes | Boolean |
- | Unified. A serialized Rust boolean. |
sample_age_ms |
yes | NonNegativeInteger |
- | Sample age ms. A non-negative Rust integer represented exactly as a JSON number. |
sample_max_age_ms |
yes | NonNegativeInteger |
- | Sample max age ms. A non-negative Rust integer represented exactly as a JSON number. |
sample_stale |
yes | Boolean |
- | Sample stale. A serialized Rust boolean. |
sampler_required |
yes | Boolean |
- | Sampler required. A serialized Rust boolean. |
sampler_running |
yes | Boolean |
- | Sampler running. A serialized Rust boolean. |
sampler_healthy |
yes | Boolean |
- | Sampler healthy. A serialized Rust boolean. |
host_backed |
yes | LiveMemoryTier | null |
- | Host backed. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "LiveMemory"
}
LiveMemoryConfigLive memory view with effective-capacity and governor-floor projections.21 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
total_bytes |
yes | NonNegativeInteger |
- | Total bytes. A non-negative Rust integer represented exactly as a JSON number. |
used_bytes |
yes | NonNegativeInteger |
- | Used bytes. A non-negative Rust integer represented exactly as a JSON number. |
available_bytes |
yes | NonNegativeInteger |
- | Available bytes. A non-negative Rust integer represented exactly as a JSON number. |
effective_capacity_available_bytes |
yes | NonNegativeInteger |
- | Effective capacity available bytes. A non-negative Rust integer represented exactly as a JSON number. |
usable_after_governor_floor_bytes |
yes | NonNegativeInteger |
- | Usable after governor floor bytes. A non-negative Rust integer represented exactly as a JSON number. |
soft_reserved_bytes |
yes | NonNegativeInteger |
- | Soft reserved bytes. A non-negative Rust integer represented exactly as a JSON number. |
sample_age_ms |
yes | NonNegativeInteger |
- | Sample age ms. A non-negative Rust integer represented exactly as a JSON number. |
sample_max_age_ms |
yes | NonNegativeInteger |
- | Sample max age ms. A non-negative Rust integer represented exactly as a JSON number. |
total_gib |
yes | NonNegativeNumber |
- | Total gib. A finite non-negative Rust floating-point value. |
used_gib |
yes | NonNegativeNumber |
- | Used gib. A finite non-negative Rust floating-point value. |
available_gib |
yes | NonNegativeNumber |
- | Available gib. A finite non-negative Rust floating-point value. |
effective_capacity_available_gib |
yes | NonNegativeNumber |
- | Effective capacity available gib. A finite non-negative Rust floating-point value. |
usable_after_governor_floor_gib |
yes | NonNegativeNumber |
- | Usable after governor floor gib. A finite non-negative Rust floating-point value. |
soft_reserved_gib |
yes | NonNegativeNumber |
- | Soft reserved gib. A finite non-negative Rust floating-point value. |
pressure |
yes | String |
- | Pressure. A serialized UTF-8 Rust string. |
source |
yes | String |
- | Source. A serialized UTF-8 Rust string. |
sample_stale |
yes | Boolean |
- | Sample stale. A serialized Rust boolean. |
sampler_required |
yes | Boolean |
- | Sampler required. A serialized Rust boolean. |
sampler_running |
yes | Boolean |
- | Sampler running. A serialized Rust boolean. |
sampler_healthy |
yes | Boolean |
- | Sampler healthy. A serialized Rust boolean. |
raw_observations |
yes | RawMemoryObservations |
- | Raw observations. Unmodified driver, host, cgroup, reserve, and host-backed memory observations. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "LiveMemoryConfig"
}
LiveMemoryTierOne independently constrained live memory tier.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
total_gb |
yes | NonNegativeNumber |
- | Total gb. A finite non-negative Rust floating-point value. |
used_gb |
yes | NonNegativeNumber |
- | Used gb. A finite non-negative Rust floating-point value. |
free_gb |
yes | NonNegativeNumber |
- | Free gb. A finite non-negative Rust floating-point value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "LiveMemoryTier"
}
LoadedAdapterDebugStateOne loaded adapter's name, path, and optional content identity.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
name |
yes | String |
- | Name. A serialized UTF-8 Rust string. |
path |
yes | String |
- | Path. A serialized UTF-8 Rust string. |
adapter_model_sha256 |
yes | Sha256 | null |
- | Adapter model SHA-256. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "LoadedAdapterDebugState"
}
LocalCapabilityAccessThe immutable access policy for an arbitrary-code-execution-grade local capability.0 fields
Type: string. Constraints: enum "loopback_only", "enabled", "disabled".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "config::LocalCapabilityAccess"
}
MaxDecodeBatchDiagnosticsConfigured, backend, and effective concurrent decode width.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | NonNegativeInteger | null |
- | Configured. |
configured_source |
yes | ConfigValueSource |
- | Configured source. The startup authority that supplied a resolved configuration value. |
backend_policy |
yes | NonNegativeInteger |
- | Backend policy. A non-negative Rust integer represented exactly as a JSON number. |
effective |
yes | NonNegativeInteger |
- | Effective. A non-negative Rust integer represented exactly as a JSON number. |
effective_source |
yes | DecodeBatchEffectiveSource |
- | Effective source. The authority that selected the effective concurrent decode width. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "MaxDecodeBatchDiagnostics"
}
MemoryBudgetConfigStartup model, KV-cache, training, and inference memory budget.9 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
total_vram_bytes |
yes | NonNegativeInteger |
- | Total VRAM bytes. A non-negative Rust integer represented exactly as a JSON number. |
model_bytes |
yes | NonNegativeInteger |
- | Model bytes. A non-negative Rust integer represented exactly as a JSON number. |
kv_cache_bytes |
yes | NonNegativeInteger |
- | KV cache bytes. A non-negative Rust integer represented exactly as a JSON number. |
training_budget_bytes |
yes | NonNegativeInteger |
- | Training budget bytes. A non-negative Rust integer represented exactly as a JSON number. |
total_vram_gib |
yes | NonNegativeNumber |
- | Total VRAM gib. A finite non-negative Rust floating-point value. |
model_gib |
yes | NonNegativeNumber |
- | Model gib. A finite non-negative Rust floating-point value. |
kv_cache_gib |
yes | NonNegativeNumber |
- | KV cache gib. A finite non-negative Rust floating-point value. |
training_budget_gib |
yes | NonNegativeNumber |
- | Training budget gib. A finite non-negative Rust floating-point value. |
inference_memory_fraction |
yes | NonNegativeNumber |
- | Inference memory fraction. A finite non-negative Rust floating-point value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "MemoryBudgetConfig"
}
MemoryGovernorConfigResolved memory-governor floor, capacity, probe, and reclaim state.9 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
floor_bytes |
yes | NonNegativeInteger |
- | Floor bytes. A non-negative Rust integer represented exactly as a JSON number. |
floor_gib |
yes | NonNegativeNumber |
- | Floor gib. A finite non-negative Rust floating-point value. |
capacity_limit_bytes |
yes | NonNegativeInteger |
- | Capacity limit bytes. A non-negative Rust integer represented exactly as a JSON number. |
capacity_limit_gib |
yes | NonNegativeNumber |
- | Capacity limit gib. A finite non-negative Rust floating-point value. |
probe_ms |
yes | NonNegativeInteger |
- | Probe ms. A non-negative Rust integer represented exactly as a JSON number. |
reclaim_mode_requested |
yes | String |
- | Reclaim mode requested. A serialized UTF-8 Rust string. |
reclaim_mode_effective |
yes | String |
- | Reclaim mode effective. A serialized UTF-8 Rust string. |
reclaim_mode_source |
yes | ConfigValueSource |
- | Reclaim mode source. The startup authority that supplied a resolved configuration value. |
reclaim_disabled_by_serving_profile |
yes | Boolean |
- | Reclaim disabled by serving profile. A serialized Rust boolean. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "MemoryGovernorConfig"
}
MemoryGovernorRuntimeInfoLive memory-governor mode, sampler health, and automatic reclaim counters.21 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
sample_age_ms |
yes | NonNegativeInteger |
- | Sample age ms. A non-negative Rust integer represented exactly as a JSON number. |
sample_max_age_ms |
yes | NonNegativeInteger |
- | Sample max age ms. A non-negative Rust integer represented exactly as a JSON number. |
automatic_attempts |
yes | NonNegativeInteger |
- | Automatic attempts. A non-negative Rust integer represented exactly as a JSON number. |
automatic_successful_attempts |
yes | NonNegativeInteger |
- | Automatic successful attempts. A non-negative Rust integer represented exactly as a JSON number. |
automatic_zero_yield_attempts |
yes | NonNegativeInteger |
- | Automatic zero yield attempts. A non-negative Rust integer represented exactly as a JSON number. |
automatic_suppressed_attempts |
yes | NonNegativeInteger |
- | Automatic suppressed attempts. A non-negative Rust integer represented exactly as a JSON number. |
automatic_reclaimed_bytes |
yes | NonNegativeInteger |
- | Automatic reclaimed bytes. A non-negative Rust integer represented exactly as a JSON number. |
automatic_last_target_bytes |
yes | NonNegativeInteger |
- | Automatic last target bytes. A non-negative Rust integer represented exactly as a JSON number. |
automatic_last_reclaimed_bytes |
yes | NonNegativeInteger |
- | Automatic last reclaimed bytes. A non-negative Rust integer represented exactly as a JSON number. |
automatic_last_duration_us |
yes | NonNegativeInteger |
- | Automatic last duration us. A non-negative Rust integer represented exactly as a JSON number. |
automatic_retry_after_ms |
yes | NonNegativeInteger |
- | Automatic retry after ms. A non-negative Rust integer represented exactly as a JSON number. |
automatic_zero_yield_streak |
yes | NonNegativeInteger |
- | Automatic zero yield streak. A non-negative Rust integer represented exactly as a JSON number. |
reclaim_mode |
yes | String |
- | Reclaim mode. A serialized UTF-8 Rust string. |
requested_reclaim_mode |
yes | String |
- | Requested reclaim mode. A serialized UTF-8 Rust string. |
automatic_monitor_enabled |
yes | Boolean |
- | Automatic monitor enabled. A serialized Rust boolean. |
sampler_required |
yes | Boolean |
- | Sampler required. A serialized Rust boolean. |
sampler_running |
yes | Boolean |
- | Sampler running. A serialized Rust boolean. |
sampler_healthy |
yes | Boolean |
- | Sampler healthy. A serialized Rust boolean. |
sample_stale |
yes | Boolean |
- | Sample stale. A serialized Rust boolean. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
disabled_by_serving_profile |
yes | Boolean |
- | Disabled by serving profile. A serialized Rust boolean. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "MemoryGovernorRuntimeInfo"
}
MemoryTierConfigRaw byte and GiB values for one independently constrained memory tier.6 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
total_bytes |
yes | NonNegativeInteger |
- | Total bytes. A non-negative Rust integer represented exactly as a JSON number. |
used_bytes |
yes | NonNegativeInteger |
- | Used bytes. A non-negative Rust integer represented exactly as a JSON number. |
free_bytes |
yes | NonNegativeInteger |
- | Free bytes. A non-negative Rust integer represented exactly as a JSON number. |
total_gib |
yes | NonNegativeNumber |
- | Total gib. A finite non-negative Rust floating-point value. |
used_gib |
yes | NonNegativeNumber |
- | Used gib. A finite non-negative Rust floating-point value. |
free_gib |
yes | NonNegativeNumber |
- | Free gib. A finite non-negative Rust floating-point value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "MemoryTierConfig"
}
MetalKernelProfileThe process-lifetime Metal backend-kernel route set.0 fields
Type: string. Constraints: enum "native_default", "portable_fallback".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "MetalKernelProfile"
}
ModelDebugStateLoaded model shape, provenance, defaults, and backend health state.10 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
path |
yes | String | null |
- | Path. |
served_model_id |
yes | String |
- | Served model ID. A serialized UTF-8 Rust string. |
defaults_profile |
yes | ModelDefaultsProfile |
- | Defaults profile. Built-in runtime defaults for the supported model family. |
num_layers |
yes | NonNegativeInteger |
- | Num layers. A non-negative Rust integer represented exactly as a JSON number. |
num_attention_heads |
yes | NonNegativeInteger |
- | Num attention heads. A non-negative Rust integer represented exactly as a JSON number. |
num_kv_heads |
yes | NonNegativeInteger |
- | Num KV heads. A non-negative Rust integer represented exactly as a JSON number. |
max_position_embeddings |
yes | NonNegativeInteger |
- | Max position embeddings. A non-negative Rust integer represented exactly as a JSON number. |
base_weight_shard_manifest |
yes | BaseWeightShardManifest | null |
- | Base weight shard manifest. |
execution_provenance |
yes | ExecutionProvenanceV1 | null |
- | Execution provenance. |
backend_runtime |
yes | BackendRuntimeInfo |
- | Backend runtime. Backend health, quarantine, restart, and external-yield settlement state. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ModelDebugState"
}
ModelDefaultsProfileBuilt-in runtime defaults for the supported model family.10 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
name |
yes | NonEmptyString |
- | Name. A non-empty serialized UTF-8 Rust string. |
canonical_model_id |
yes | NonEmptyString |
- | Canonical model ID. A non-empty serialized UTF-8 Rust string. |
canonical_served_model_id |
yes | NonEmptyString |
- | Canonical served model ID. A non-empty serialized UTF-8 Rust string. |
server_default_thinking_enabled |
yes | Boolean | null |
- | Server default thinking enabled. |
template_default_thinking_enabled |
yes | Boolean |
- | Template default thinking enabled. A serialized Rust boolean. |
eval_mode_default_thinking_enabled |
yes | Boolean |
- | Eval mode default thinking enabled. A serialized Rust boolean. |
adapter_dir_policy |
yes | String |
- | Adapter dir policy. A serialized UTF-8 Rust string. |
chat_template_policy |
yes | String |
- | Chat template policy. A serialized UTF-8 Rust string. |
supports_enable_thinking_kwarg |
yes | Boolean |
- | Supports enable thinking kwarg. A serialized Rust boolean. |
supports_tool_chat_template |
yes | Boolean |
- | Supports tool chat template. A serialized Rust boolean. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ModelDefaultsProfile"
}
ModelInfoOne OpenAI-compatible served model descriptor.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
id |
yes | String |
- | Served model ID accepted by the inference API. |
object |
yes | any |
const "model" | OpenAI-compatible model discriminator. |
owned_by |
yes | any |
const "kiln" | Stable owner label for a model served by Kiln. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ModelInfo"
}
ModelStartupConfigResponseResolved model-startup resource policy and observations.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
checkpoint_read |
yes | CheckpointReadConfigResponse |
- | Checkpoint read. Resolved startup-only checkpoint-read policy and completed real-model observations. |
accelerator_weight_upload |
yes | AcceleratorWeightUploadConfigResponse |
- | Accelerator weight upload. Resolved startup-only accelerator-weight upload policy and completed real-model observation. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ModelStartupConfigResponse"
}
ModelStateResponseComplete opt-in model, adapter, provenance, typed runtime, and cache debug state.19 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
model |
yes | ModelDebugState |
- | Model. Loaded model shape, provenance, defaults, and backend health state. |
adapters |
yes | AdapterDebugState |
- | Adapters. Active, loaded, discoverable, and failed adapter state. |
config_hashes |
yes | ConfigHashes |
- | Config hashes. Available hashes of loaded model, tokenizer, template, and effective environment configuration. |
http |
yes | HttpRuntimeInfo |
- | HTTP. Resolved and kernel-readback HTTP socket send-buffer state. |
decode_runtime |
yes | DecodeRuntimeConfig |
- | Decode runtime. The immutable process-lifetime decode policy. |
accelerator_runtime |
yes | ResolvedAcceleratorRuntimePolicy |
- | Accelerator runtime. Versioned process-lifetime accelerator policy. |
cuda_graphs |
yes | CudaGraphConfigResponse |
- | CUDA graphs. Startup-resolved CUDA graph request, cache bound, and fixed safety invariants. |
cuda_synchronization |
yes | CudaSynchronizationRuntimeStats |
- | CUDA synchronization. Point-in-time CUDA synchronization telemetry. |
rocm_synchronization |
yes | RocmSynchronizationRuntimeStats |
- | ROCm synchronization. Point-in-time ROCm synchronization telemetry. |
rocm_graphs |
yes | RocmGraphStats | null |
- | Point-in-time ROCm graph statistics, or null when a nonblocking snapshot is unavailable. |
rocm_graphs_unavailable_reason |
yes | RocmGraphUnavailableReason | null |
- | Closed reason rocm_graphs is null; null when statistics are present. |
rocm_graph_telemetry |
yes | RocmGraphLiveTelemetry | null |
- | Lock-independent ROCm graph phase telemetry, or null when no graph runner exists. |
rocm_graph_telemetry_unavailable_reason |
yes | RocmGraphUnavailableReason | null |
- | Closed reason rocm_graph_telemetry is null; null when telemetry is present. |
kv_autoscaler |
yes | KvAutoscalerState |
- | KV autoscaler. Startup and activation state for the physical KV-cache autoscaler. |
streaming_prefill |
yes | StreamingPrefillRuntimeConfig |
- | Streaming prefill. Complete operator-facing resolved streaming-prefill policy. |
training |
yes | TrainingDebugState |
- | Training. Training checkpoint-boundary state published by the debug endpoint. |
batching_engine |
yes | BatchingEngineDebugState |
- | Batching engine. Selected batching backend, resolved configuration, and live execution state. |
thinking |
yes | ThinkingDebugState |
- | Thinking. Resolved thinking defaults from request-independent startup policy. |
caches |
yes | CacheDebugState |
- | Caches. Deterministic response, prompt, token, prefix-cache, and backend recurrent-state ownership. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ModelStateResponse"
}
ModelsResponseOpenAI-compatible list of served models.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
object |
yes | any |
const "list" | OpenAI-compatible list discriminator. |
data |
yes | array<ModelInfo> |
minimum items 1; maximum items 1 | The single model served by this Kiln process. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ModelsResponse"
}
NonEmptyStringA non-empty serialized UTF-8 Rust string.0 fields
Type: string. Constraints: minimum length 1.
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "String"
}
NonNegativeIntegerA non-negative Rust integer represented exactly as a JSON number.0 fields
Type: integer. Constraints: minimum 0.
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "u64 | u32 | usize"
}
NonNegativeNumberA finite non-negative Rust floating-point value.0 fields
Type: number. Constraints: minimum 0.
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "f32 | f64"
}
OperationalRuntimeConfigStartup-resolved operational paths and local-capability access policy retained by request handlers.9 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
bind_host |
yes | String |
- | Bind host. A serialized UTF-8 Rust string. |
terminal_access |
yes | LocalCapabilityAccess |
- | Terminal access. The immutable access policy for an arbitrary-code-execution-grade local capability. |
terminal_enabled |
yes | Boolean |
- | Terminal enabled. A serialized Rust boolean. |
agent_runs_access |
yes | LocalCapabilityAccess |
- | Agent runs access. The immutable access policy for an arbitrary-code-execution-grade local capability. |
agent_runs_enabled |
yes | Boolean |
- | Agent runs enabled. A serialized Rust boolean. |
pi_bin |
yes | String | null |
- | Pi bin. |
pi_sessions_dir |
yes | String |
- | Pi sessions dir. A serialized UTF-8 Rust string. |
adapter_library_url |
yes | String |
- | Adapter library URL. A serialized UTF-8 Rust string. |
logit_cache_dir |
yes | String |
- | Logit cache dir. A serialized UTF-8 Rust string. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "config::OperationalRuntimeConfig"
}
PositiveIntegerA positive Rust integer represented exactly as a JSON number.0 fields
Type: integer. Constraints: minimum 1.
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "NonZeroUsize"
}
PrefillAdmissionQuantumDiagnosticsConfigured, backend, and effective prefill admission quantum.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | NonNegativeInteger | null |
- | Configured. |
configured_source |
yes | ConfigValueSource |
- | Configured source. The startup authority that supplied a resolved configuration value. |
backend_policy |
yes | NonNegativeInteger |
- | Backend policy. A non-negative Rust integer represented exactly as a JSON number. |
effective |
yes | NonNegativeInteger |
- | Effective. A non-negative Rust integer represented exactly as a JSON number. |
effective_source |
yes | BatchingEffectiveSource |
- | Effective source. The authority that selected an effective batching value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "PrefillAdmissionQuantumDiagnostics"
}
PrefillRuntimeInfoResolved prefill execution state.1 field
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
streaming_prefill |
yes | StreamingPrefillRuntimeConfig |
- | Streaming prefill. Complete operator-facing resolved streaming-prefill policy. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "PrefillRuntimeInfo"
}
PrefixCacheConfigRequested startup prefix-cache policy.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
enabled |
yes | Boolean |
- | Enabled. A serialized Rust boolean. |
max_blocks |
yes | PositiveInteger | null |
- | Max blocks. |
max_entries |
yes | PositiveInteger | null |
- | Max entries. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "config::PrefixCacheConfig"
}
PrefixCacheConfigResponseRequested prefix-cache policy plus backend-qualified live limits.6 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configuration |
yes | PrefixCacheConfig |
- | Configuration. Requested startup prefix-cache policy. |
effective_enabled |
yes | Boolean |
- | Effective enabled. A serialized Rust boolean. |
effective_reason |
yes | String |
- | Effective reason. A serialized UTF-8 Rust string. |
effective_max_blocks |
yes | NonNegativeInteger |
- | Effective max blocks. A non-negative Rust integer represented exactly as a JSON number. |
effective_max_entries |
yes | NonNegativeInteger |
- | Effective max entries. A non-negative Rust integer represented exactly as a JSON number. |
effective_max_state_bytes |
yes | NonNegativeInteger |
- | Effective max state bytes. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "PrefixCacheConfigResponse"
}
PrefixCacheHealthInfoPrefix-cache counters plus normalized utilization for health reporting.16 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
enabled |
yes | Boolean |
- | Enabled. A serialized Rust boolean. |
lookup_hits |
yes | NonNegativeInteger |
- | Lookup hits. A non-negative Rust integer represented exactly as a JSON number. |
lookup_misses |
yes | NonNegativeInteger |
- | Lookup misses. A non-negative Rust integer represented exactly as a JSON number. |
hit_tokens |
yes | NonNegativeInteger |
- | Hit tokens. A non-negative Rust integer represented exactly as a JSON number. |
hit_blocks |
yes | NonNegativeInteger |
- | Hit blocks. A non-negative Rust integer represented exactly as a JSON number. |
cached_blocks |
yes | NonNegativeInteger |
- | Cached blocks. A non-negative Rust integer represented exactly as a JSON number. |
max_blocks |
yes | NonNegativeInteger |
- | Max blocks. A non-negative Rust integer represented exactly as a JSON number. |
cached_entries |
yes | NonNegativeInteger |
- | Cached entries. A non-negative Rust integer represented exactly as a JSON number. |
max_entries |
yes | NonNegativeInteger |
- | Max entries. A non-negative Rust integer represented exactly as a JSON number. |
cached_state_bytes |
yes | NonNegativeInteger |
- | Cached state bytes. A non-negative Rust integer represented exactly as a JSON number. |
max_state_bytes |
yes | NonNegativeInteger |
- | Max state bytes. A non-negative Rust integer represented exactly as a JSON number. |
active_leases |
yes | NonNegativeInteger |
- | Active leases. A non-negative Rust integer represented exactly as a JSON number. |
pending_release_entries |
yes | NonNegativeInteger |
- | Pending release entries. A non-negative Rust integer represented exactly as a JSON number. |
block_utilization |
yes | NonNegativeNumber |
- | Block utilization. A finite non-negative Rust floating-point value. |
entry_utilization |
yes | NonNegativeNumber |
- | Entry utilization. A finite non-negative Rust floating-point value. |
state_utilization |
yes | NonNegativeNumber |
- | State utilization. A finite non-negative Rust floating-point value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "PrefixCacheInfo"
}
PrefixCacheInfoPrefix-cache occupancy, hit, limit, and lease counters.13 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
enabled |
yes | Boolean |
- | Enabled. A serialized Rust boolean. |
lookup_hits |
yes | NonNegativeInteger |
- | Lookup hits. A non-negative Rust integer represented exactly as a JSON number. |
lookup_misses |
yes | NonNegativeInteger |
- | Lookup misses. A non-negative Rust integer represented exactly as a JSON number. |
hit_tokens |
yes | NonNegativeInteger |
- | Hit tokens. A non-negative Rust integer represented exactly as a JSON number. |
hit_blocks |
yes | NonNegativeInteger |
- | Hit blocks. A non-negative Rust integer represented exactly as a JSON number. |
cached_blocks |
yes | NonNegativeInteger |
- | Cached blocks. A non-negative Rust integer represented exactly as a JSON number. |
max_blocks |
yes | NonNegativeInteger |
- | Max blocks. A non-negative Rust integer represented exactly as a JSON number. |
cached_entries |
yes | NonNegativeInteger |
- | Cached entries. A non-negative Rust integer represented exactly as a JSON number. |
max_entries |
yes | NonNegativeInteger |
- | Max entries. A non-negative Rust integer represented exactly as a JSON number. |
cached_state_bytes |
yes | NonNegativeInteger |
- | Cached state bytes. A non-negative Rust integer represented exactly as a JSON number. |
max_state_bytes |
yes | NonNegativeInteger |
- | Max state bytes. A non-negative Rust integer represented exactly as a JSON number. |
active_leases |
yes | NonNegativeInteger |
- | Active leases. A non-negative Rust integer represented exactly as a JSON number. |
pending_release_entries |
yes | NonNegativeInteger |
- | Pending release entries. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "PrefixCacheInfo | PrefixCacheDebugState"
}
PromptCacheInfoOne bounded prompt-cache hit and occupancy snapshot.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
hits |
yes | NonNegativeInteger |
- | Hits. A non-negative Rust integer represented exactly as a JSON number. |
misses |
yes | NonNegativeInteger |
- | Misses. A non-negative Rust integer represented exactly as a JSON number. |
entries |
yes | NonNegativeInteger |
- | Entries. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "PromptCacheInfo | PromptCacheDebugState"
}
PromptCachesInfoRendered-prompt and prompt-token cache snapshots.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
rendered_prompt |
yes | PromptCacheInfo |
- | Rendered prompt. One bounded prompt-cache hit and occupancy snapshot. |
prompt_token |
yes | PromptCacheInfo |
- | Prompt token. One bounded prompt-cache hit and occupancy snapshot. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "PromptCachesInfo"
}
RawMemoryObservationsUnmodified driver, host, cgroup, reserve, and host-backed memory observations.16 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
probe_failed |
yes | Boolean |
- | Probe failed. A serialized Rust boolean. |
driver_total_bytes |
yes | NonNegativeInteger | null |
- | Driver total bytes. |
driver_used_bytes |
yes | NonNegativeInteger | null |
- | Driver used bytes. |
driver_free_bytes |
yes | NonNegativeInteger | null |
- | Driver free bytes. |
driver_vram_total_bytes |
yes | NonNegativeInteger | null |
- | Driver VRAM total bytes. |
driver_vram_used_bytes |
yes | NonNegativeInteger | null |
- | Driver VRAM used bytes. |
driver_gtt_total_bytes |
yes | NonNegativeInteger | null |
- | Driver gtt total bytes. |
driver_gtt_used_bytes |
yes | NonNegativeInteger | null |
- | Driver gtt used bytes. |
host_total_bytes |
yes | NonNegativeInteger | null |
- | Host total bytes. |
host_available_bytes |
yes | NonNegativeInteger | null |
- | Host available bytes. |
cgroup_limit_bytes |
yes | NonNegativeInteger | null |
- | Cgroup limit bytes. |
cgroup_high_bytes |
yes | NonNegativeInteger | null |
- | Cgroup high bytes. |
cgroup_current_bytes |
yes | NonNegativeInteger | null |
- | Cgroup current bytes. |
cgroup_remaining_bytes |
yes | NonNegativeInteger | null |
- | Cgroup remaining bytes. |
unified_reserve_bytes |
yes | NonNegativeInteger | null |
- | Unified reserve bytes. |
host_backed |
yes | MemoryTierConfig | null |
- | Host backed. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RawMemoryObservations"
}
RecentRequestMetricsAggregated metrics over the bounded recent-request ring.7 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
retained |
yes | NonNegativeInteger |
- | Retained. A non-negative Rust integer represented exactly as a JSON number. |
capacity |
yes | NonNegativeInteger |
- | Capacity. A non-negative Rust integer represented exactly as a JSON number. |
latency_ms |
yes | LatencyPercentiles |
- | Latency ms. Latency percentiles in milliseconds. |
tokens_per_second |
yes | NonNegativeNumber |
- | Tokens per second. A finite non-negative Rust floating-point value. |
timeout_count |
yes | NonNegativeInteger |
- | Timeout count. A non-negative Rust integer represented exactly as a JSON number. |
error_count |
yes | NonNegativeInteger |
- | Error count. A non-negative Rust integer represented exactly as a JSON number. |
last_error |
yes | LastErrorSummary | null |
- | Last error. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RecentRequestMetrics"
}
RequestLatencyDiagnosticsRequest-local TTFT, ITL tail, bounded stall reasons, and honest phase-coverage diagnostics.14 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
emitted_tokens |
yes | NonNegativeInteger |
- | Emitted tokens. A non-negative Rust integer represented exactly as a JSON number. |
gap_samples |
yes | NonNegativeInteger |
- | Gap samples. A non-negative Rust integer represented exactly as a JSON number. |
retained_gap_samples |
yes | integer |
minimum 0; maximum 8192 | Retained gap samples. |
gap_samples_truncated |
yes | Boolean |
- | Gap samples truncated. A serialized Rust boolean. |
ttft_ms |
yes | NonNegativeNumber | null |
- | TTFT ms. |
itl_ms_p50 |
yes | NonNegativeNumber | null |
- | ITL ms p50. |
itl_ms_p99 |
yes | NonNegativeNumber | null |
- | ITL ms p99. |
itl_ms_p999 |
yes | NonNegativeNumber | null |
- | ITL ms p999. |
max_itl_ms |
yes | NonNegativeNumber | null |
- | Max ITL ms. |
stall_threshold_ms |
yes | NonNegativeNumber | null |
- | Stall threshold ms. |
stall_count |
yes | NonNegativeInteger |
- | Stall count. A non-negative Rust integer represented exactly as a JSON number. |
unexplained_stall_count |
yes | NonNegativeInteger |
- | Unexplained stall count. A non-negative Rust integer represented exactly as a JSON number. |
stall_reasons |
yes | LatencyStallReasonCounts |
- | Stall reasons. Fixed-cardinality token-stall counts by dominant blocking reason. |
phases |
yes | LatencyPhaseTimings |
- | Phases. Bounded request phase timings; null means the serving path did not measure that subphase. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RequestLatencyDiagnostics"
}
RequestMetricsLifetime request outcome and concurrency counters.7 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
total |
yes | NonNegativeInteger |
- | Total. A non-negative Rust integer represented exactly as a JSON number. |
ok |
yes | NonNegativeInteger |
- | Ok. A non-negative Rust integer represented exactly as a JSON number. |
error |
yes | NonNegativeInteger |
- | Error. A non-negative Rust integer represented exactly as a JSON number. |
timeout |
yes | NonNegativeInteger |
- | Timeout. A non-negative Rust integer represented exactly as a JSON number. |
rejected |
yes | NonNegativeInteger |
- | Rejected. A non-negative Rust integer represented exactly as a JSON number. |
active |
yes | NonNegativeInteger |
- | Active. A non-negative Rust integer represented exactly as a JSON number. |
active_peak |
yes | NonNegativeInteger |
- | Active peak. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RequestMetrics"
}
RequestRecordOne newest-first bounded recent-request record. The record can contain full prompt and completion text and therefore belongs inside the server's trusted access boundary.26 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
id |
yes | String |
- | ID. A serialized UTF-8 Rust string. |
timestamp_unix_ms |
yes | NonNegativeInteger |
- | Timestamp unix ms. A non-negative Rust integer represented exactly as a JSON number. |
model |
yes | String |
- | Model. A serialized UTF-8 Rust string. |
prompt_preview |
yes | String |
- | Prompt preview. A serialized UTF-8 Rust string. |
completion_preview |
yes | String |
- | Completion preview. A serialized UTF-8 Rust string. |
prompt_tokens |
yes | NonNegativeInteger |
- | Prompt tokens. A non-negative Rust integer represented exactly as a JSON number. |
completion_tokens |
yes | NonNegativeInteger |
- | Completion tokens. A non-negative Rust integer represented exactly as a JSON number. |
duration_ms |
yes | NonNegativeInteger |
- | Duration ms. A non-negative Rust integer represented exactly as a JSON number. |
streamed |
yes | Boolean |
- | Streamed. A serialized Rust boolean. |
finish_reason |
yes | String |
- | Finish reason. A serialized UTF-8 Rust string. |
adapter |
no | String |
- | Adapter. A serialized UTF-8 Rust string. |
temperature |
no | FiniteNumber |
- | Temperature. A finite Rust floating-point value. serde_json rejects non-finite values. |
top_p |
no | FiniteNumber |
- | Top p. A finite Rust floating-point value. serde_json rejects non-finite values. |
max_tokens |
no | NonNegativeInteger |
- | Max tokens. A non-negative Rust integer represented exactly as a JSON number. |
ttft_ms |
no | NonNegativeInteger |
- | TTFT ms. A non-negative Rust integer represented exactly as a JSON number. |
model_prefill_ms |
no | NonNegativeInteger |
- | Model prefill ms. A non-negative Rust integer represented exactly as a JSON number. |
model_decode_ms |
no | NonNegativeInteger |
- | Model decode ms. A non-negative Rust integer represented exactly as a JSON number. |
error |
no | String |
- | Error. A serialized UTF-8 Rust string. |
thinking_mode |
no | String |
- | Thinking mode. A serialized UTF-8 Rust string. |
prefix_cache |
no | String |
- | Prefix cache. A serialized UTF-8 Rust string. |
prompt_full |
no | String |
- | Prompt full. A serialized UTF-8 Rust string. |
completion_full |
no | String |
- | Completion full. A serialized UTF-8 Rust string. |
user_agent |
no | String |
- | User agent. A serialized UTF-8 Rust string. |
client |
no | String |
- | Client. A serialized UTF-8 Rust string. |
thinking_budget |
no | RequestThinkingBudget |
- | Thinking budget. Configured thinking budget and observed request outcome. |
latency |
no | RequestLatencyDiagnostics |
- | Latency. Request-local TTFT, ITL tail, bounded stall reasons, and honest phase-coverage diagnostics. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RequestRecord"
}
RequestThinkingBudgetConfigured thinking budget and observed request outcome.11 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | Boolean |
- | Configured. A serialized Rust boolean. |
max_tokens |
no | NonNegativeInteger |
- | Max tokens. A non-negative Rust integer represented exactly as a JSON number. |
max_time_ms |
no | NonNegativeInteger |
- | Max time ms. A non-negative Rust integer represented exactly as a JSON number. |
tokens_source |
yes | ThinkingBudgetSource |
- | Tokens source. The stable origin of an effective thinking-budget value. |
time_source |
yes | ThinkingBudgetSource |
- | Time source. The stable origin of an effective thinking-budget value. |
applied |
no | Boolean |
- | Applied. A serialized Rust boolean. |
triggered |
no | Boolean |
- | Triggered. A serialized Rust boolean. |
trigger |
no | String |
- | Trigger. A serialized UTF-8 Rust string. |
closed |
no | Boolean |
- | Closed. A serialized Rust boolean. |
thinking_tokens |
no | NonNegativeInteger |
- | Thinking tokens. A non-negative Rust integer represented exactly as a JSON number. |
thinking_time_ms |
no | NonNegativeInteger |
- | Thinking time ms. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RequestThinkingBudget"
}
ResolvedAcceleratorBooleanOne configured/effective accelerator policy leaf and its startup source.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | Boolean |
- | Configured. A serialized Rust boolean. |
effective |
yes | Boolean |
- | Effective. A serialized Rust boolean. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ResolvedAcceleratorValue<bool>"
}
ResolvedAcceleratorIntegerOne configured/effective accelerator policy leaf and its startup source.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | NonNegativeInteger |
- | Configured. A non-negative Rust integer represented exactly as a JSON number. |
effective |
yes | NonNegativeInteger |
- | Effective. A non-negative Rust integer represented exactly as a JSON number. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ResolvedAcceleratorValue<usize | u64>"
}
ResolvedAcceleratorOptionalIntegerOne configured/effective accelerator policy leaf and its startup source.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | NonNegativeInteger | null |
- | Configured. |
effective |
yes | NonNegativeInteger | null |
- | Effective. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ResolvedAcceleratorValue<Option<usize>>"
}
ResolvedAcceleratorRuntimePolicyVersioned process-lifetime accelerator policy.21 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
schema_id |
yes | NonEmptyString |
- | Schema ID. A non-empty serialized UTF-8 Rust string. |
version |
yes | NonNegativeInteger |
- | Version. A non-negative Rust integer represented exactly as a JSON number. |
vulkan_kernel_policy_schema_id |
yes | NonEmptyString |
- | Vulkan kernel policy schema ID. A non-empty serialized UTF-8 Rust string. |
vulkan_device_policy_schema_id |
yes | NonEmptyString |
- | Vulkan device policy schema ID. A non-empty serialized UTF-8 Rust string. |
serving_profile |
yes | ServingProfile |
- | Serving profile. The immutable process-lifetime serving profile. |
serving_profile_source |
yes | ConfigValueSource |
- | Serving profile source. The startup authority that supplied a resolved configuration value. |
kt_api_mode |
yes | ResolvedKtApiMode |
- | Kt API mode. One configured/effective accelerator policy leaf and its startup source. |
full_attention_score_budget_mib |
yes | ResolvedAcceleratorInteger |
- | Full attention score budget MiB. One configured/effective accelerator policy leaf and its startup source. |
vulkan_device_index |
yes | ResolvedAcceleratorOptionalInteger |
- | Vulkan device index. One configured/effective accelerator policy leaf and its startup source. |
vulkan_validation |
yes | ResolvedAcceleratorBoolean |
- | Vulkan validation. One configured/effective accelerator policy leaf and its startup source. |
cuda_kernel_profile |
yes | ResolvedCudaKernelProfile |
- | CUDA kernel profile. One configured/effective accelerator policy leaf and its startup source. |
cuda_marlin_profile |
yes | ResolvedCudaMarlinProfile |
- | CUDA marlin profile. One configured/effective accelerator policy leaf and its startup source. |
cuda_flash_backward_mode |
yes | ResolvedCudaFlashBackwardMode |
- | CUDA flash backward mode. One configured/effective accelerator policy leaf and its startup source. |
metal_kernel_profile |
yes | ResolvedMetalKernelProfile |
- | Metal kernel profile. One configured/effective accelerator policy leaf and its startup source. |
rocm_synchronization_mode |
yes | ResolvedRocmSynchronizationMode |
- | ROCm synchronization mode. One configured/effective accelerator policy leaf and its startup source. |
rocm_strided_batched_matmul_mode |
yes | ResolvedRocmStridedBatchedMatmulMode |
- | ROCm strided batched matmul mode. One configured/effective accelerator policy leaf and its startup source. |
rocm_bf16_matmul_output_mode |
yes | ResolvedRocmBf16MatmulOutputMode |
- | ROCm BF16 matmul output mode. One configured/effective accelerator policy leaf and its startup source. |
rocm_kernel_profile |
yes | ResolvedRocmKernelProfile |
- | ROCm kernel profile. One configured/effective accelerator policy leaf and its startup source. |
rocm_graph_mode |
yes | ResolvedRocmGraphMode |
- | ROCm graph mode. One configured/effective accelerator policy leaf and its startup source. |
rocm_graph_cache_entries |
yes | ResolvedAcceleratorInteger |
- | ROCm graph cache entries. One configured/effective accelerator policy leaf and its startup source. |
rocm_graph_cache_max_bytes |
yes | ResolvedAcceleratorInteger |
- | ROCm graph cache max bytes. One configured/effective accelerator policy leaf and its startup source. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ResolvedAcceleratorRuntimePolicy"
}
ResolvedApplicationPathsAbsolute process-lifetime cache root and the startup authority that selected it.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
cache_root |
yes | String |
- | Cache root. A serialized UTF-8 Rust string. |
cache_root_source |
yes | ConfigValueSource |
- | Cache root source. The startup authority that supplied a resolved configuration value. |
restart_required_to_change |
yes | any |
const true | Restart required to change. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "config::ResolvedApplicationPaths"
}
ResolvedCudaFlashBackwardModeOne configured/effective accelerator policy leaf and its startup source.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | CudaFlashBackwardMode |
- | Configured. The process-lifetime CUDA FlashAttention backward accumulation mode. |
effective |
yes | CudaFlashBackwardMode |
- | Effective. The process-lifetime CUDA FlashAttention backward accumulation mode. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ResolvedAcceleratorValue<CudaFlashBackwardMode>"
}
ResolvedCudaKernelProfileOne configured/effective accelerator policy leaf and its startup source.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | CudaKernelProfile |
- | Configured. The process-lifetime CUDA backend-kernel route set. |
effective |
yes | CudaKernelProfile |
- | Effective. The process-lifetime CUDA backend-kernel route set. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ResolvedAcceleratorValue<CudaKernelProfile>"
}
ResolvedCudaMarlinProfileOne configured/effective accelerator policy leaf and its startup source.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | CudaMarlinProfile |
- | Configured. The process-lifetime CUDA Marlin projection layout. |
effective |
yes | CudaMarlinProfile |
- | Effective. The process-lifetime CUDA Marlin projection layout. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ResolvedAcceleratorValue<CudaMarlinProfile>"
}
ResolvedKtApiModeOne configured/effective accelerator policy leaf and its startup source.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | KtApiMode |
- | Configured. The process-lifetime kiln-tensor adapter route selection. |
effective |
yes | KtApiMode |
- | Effective. The process-lifetime kiln-tensor adapter route selection. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ResolvedAcceleratorValue<KtApiMode>"
}
ResolvedMetalKernelProfileOne configured/effective accelerator policy leaf and its startup source.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | MetalKernelProfile |
- | Configured. The process-lifetime Metal backend-kernel route set. |
effective |
yes | MetalKernelProfile |
- | Effective. The process-lifetime Metal backend-kernel route set. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ResolvedAcceleratorValue<MetalKernelProfile>"
}
ResolvedRocmBf16MatmulOutputModeOne configured/effective accelerator policy leaf and its startup source.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | RocmBf16MatmulOutputMode |
- | Configured. The process-lifetime ROCm BF16-output matmul route. |
effective |
yes | RocmBf16MatmulOutputMode |
- | Effective. The process-lifetime ROCm BF16-output matmul route. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ResolvedAcceleratorValue<RocmBf16MatmulOutputMode>"
}
ResolvedRocmGraphModeOne configured/effective accelerator policy leaf and its startup source.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | RocmGraphMode |
- | Configured. The configured ROCm graph lifecycle. |
effective |
yes | RocmGraphMode |
- | Effective. The configured ROCm graph lifecycle. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ResolvedAcceleratorValue<RocmGraphMode>"
}
ResolvedRocmKernelProfileOne configured/effective accelerator policy leaf and its startup source.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | RocmKernelProfile |
- | Configured. The complete process-lifetime ROCm model-kernel route set. |
effective |
yes | RocmKernelProfile |
- | Effective. The complete process-lifetime ROCm model-kernel route set. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ResolvedAcceleratorValue<RocmKernelProfile>"
}
ResolvedRocmStridedBatchedMatmulModeOne configured/effective accelerator policy leaf and its startup source.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | RocmStridedBatchedMatmulMode |
- | Configured. The process-lifetime ROCm strided-batched matmul route. |
effective |
yes | RocmStridedBatchedMatmulMode |
- | Effective. The process-lifetime ROCm strided-batched matmul route. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ResolvedAcceleratorValue<RocmStridedBatchedMatmulMode>"
}
ResolvedRocmSynchronizationModeOne configured/effective accelerator policy leaf and its startup source.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | RocmSynchronizationMode |
- | Configured. The process-lifetime ROCm synchronization policy. |
effective |
yes | RocmSynchronizationMode |
- | Effective. The process-lifetime ROCm synchronization policy. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ResolvedAcceleratorValue<RocmSynchronizationMode>"
}
RocmBf16MatmulOutputModeThe process-lifetime ROCm BF16-output matmul route.0 fields
Type: string. Constraints: enum "f32_then_cast", "native_bf16".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RocmBf16MatmulOutputMode"
}
RocmGraphFallbackStatsClosed-reason ROCm eager-fallback counts and end-to-end latency.18 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
total |
yes | NonNegativeInteger |
- | Total. A non-negative Rust integer represented exactly as a JSON number. |
multi_row_batch_unsupported |
yes | NonNegativeInteger |
- | Multi row batch unsupported. A non-negative Rust integer represented exactly as a JSON number. |
cold_cache_host_round_trip |
yes | NonNegativeInteger |
- | Cold cache host round trip. A non-negative Rust integer represented exactly as a JSON number. |
persistent_host_round_trip |
yes | NonNegativeInteger |
- | Persistent host round trip. A non-negative Rust integer represented exactly as a JSON number. |
shape_dependent_attention |
yes | NonNegativeInteger |
- | Shape dependent attention. A non-negative Rust integer represented exactly as a JSON number. |
graph_cache_capacity |
yes | NonNegativeInteger |
- | Graph cache capacity. A non-negative Rust integer represented exactly as a JSON number. |
graph_cache_byte_budget |
yes | NonNegativeInteger |
- | Graph cache byte budget. A non-negative Rust integer represented exactly as a JSON number. |
graph_accounting_incomplete |
yes | NonNegativeInteger |
- | Graph accounting incomplete. A non-negative Rust integer represented exactly as a JSON number. |
moderate_memory_pressure |
yes | NonNegativeInteger |
- | Moderate memory pressure. A non-negative Rust integer represented exactly as a JSON number. |
tight_memory_pressure |
yes | NonNegativeInteger |
- | Tight memory pressure. A non-negative Rust integer represented exactly as a JSON number. |
critical_memory_pressure |
yes | NonNegativeInteger |
- | Critical memory pressure. A non-negative Rust integer represented exactly as a JSON number. |
memory_reservation_denied |
yes | NonNegativeInteger |
- | Memory reservation denied. A non-negative Rust integer represented exactly as a JSON number. |
memory_governor_selector_mismatch |
yes | NonNegativeInteger |
- | Memory governor selector mismatch. A non-negative Rust integer represented exactly as a JSON number. |
capture_failure |
yes | NonNegativeInteger |
- | Capture failure. A non-negative Rust integer represented exactly as a JSON number. |
replay_failure |
yes | NonNegativeInteger |
- | Replay failure. A non-negative Rust integer represented exactly as a JSON number. |
slow |
yes | NonNegativeInteger |
- | Slow. A non-negative Rust integer represented exactly as a JSON number. |
total_duration_micros |
yes | NonNegativeInteger |
- | Total duration micros. A non-negative Rust integer represented exactly as a JSON number. |
max_duration_micros |
yes | NonNegativeInteger |
- | Max duration micros. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RocmGraphFallbackStats"
}
RocmGraphInfoNonblocking ROCm graph state projected for health reporting.62 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
requested |
yes | Boolean | null |
- | Requested. |
capture_requested |
yes | Boolean | null |
- | Capture requested. |
enabled |
yes | Boolean | null |
- | Enabled. |
capture_enabled |
yes | Boolean | null |
- | Capture enabled. |
state |
yes | string |
enum "enabled", "disabled", "busy", "unavailable" | State. |
unavailable_reason |
yes | RocmGraphUnavailableReason | null |
- | Unavailable reason. |
phase_telemetry_available |
yes | Boolean |
- | Phase telemetry available. A serialized Rust boolean. |
phase_telemetry_unavailable_reason |
yes | RocmGraphUnavailableReason | null |
- | Phase telemetry unavailable reason. |
current_phase |
yes | RocmGraphPhase | null |
- | Current phase. |
current_phase_elapsed_micros |
yes | NonNegativeInteger | null |
- | Current phase elapsed micros. |
max_cached_graphs |
yes | NonNegativeInteger | null |
- | Max cached graphs. |
max_retained_bytes |
yes | NonNegativeInteger | null |
- | Max retained bytes. |
capture_attempts |
yes | NonNegativeInteger | null |
- | Capture attempts. |
capture_successes |
yes | NonNegativeInteger | null |
- | Capture successes. |
capture_deferrals |
yes | NonNegativeInteger | null |
- | Capture deferrals. |
capture_failures |
yes | NonNegativeInteger | null |
- | Capture failures. |
replay_attempts |
yes | NonNegativeInteger | null |
- | Replay attempts. |
replay_successes |
yes | NonNegativeInteger | null |
- | Replay successes. |
replay_failures |
yes | NonNegativeInteger | null |
- | Replay failures. |
failures |
yes | NonNegativeInteger | null |
- | Failures. |
decode_owner_release_count |
yes | NonNegativeInteger | null |
- | Decode owner release count. |
decode_owner_graph_release_count |
yes | NonNegativeInteger | null |
- | Decode owner graph release count. |
graph_slot_create_count |
yes | NonNegativeInteger | null |
- | Graph slot create count. |
graph_slot_reuse_count |
yes | NonNegativeInteger | null |
- | Graph slot reuse count. |
cache_admission_successes |
yes | NonNegativeInteger | null |
- | Cache admission successes. |
cache_evictions |
yes | NonNegativeInteger | null |
- | Cache evictions. |
cache_evicted_bytes |
yes | NonNegativeInteger | null |
- | Cache evicted bytes. |
budget_evictions |
yes | NonNegativeInteger | null |
- | Budget evictions. |
pressure_evictions |
yes | NonNegativeInteger | null |
- | Pressure evictions. |
invalidation_evictions |
yes | NonNegativeInteger | null |
- | Invalidation evictions. |
recovery_evictions |
yes | NonNegativeInteger | null |
- | Recovery evictions. |
entry_capacity_rejections |
yes | NonNegativeInteger | null |
- | Entry capacity rejections. |
byte_budget_rejections |
yes | NonNegativeInteger | null |
- | Byte budget rejections. |
accounting_incomplete_rejections |
yes | NonNegativeInteger | null |
- | Accounting incomplete rejections. |
pre_capture_entry_capacity_skips |
yes | NonNegativeInteger | null |
- | Pre capture entry capacity skips. |
pre_capture_byte_budget_skips |
yes | NonNegativeInteger | null |
- | Pre capture byte budget skips. |
pre_capture_accounting_incomplete_skips |
yes | NonNegativeInteger | null |
- | Pre capture accounting incomplete skips. |
pre_capture_memory_reservation_denied_skips |
yes | NonNegativeInteger | null |
- | Pre capture memory reservation denied skips. |
memory_governor_selector_mismatch_skips |
yes | NonNegativeInteger | null |
- | Memory governor selector mismatch skips. |
captured_graph_count |
yes | NonNegativeInteger | null |
- | Captured graph count. |
graph_slot_count |
yes | NonNegativeInteger | null |
- | Graph slot count. |
active_graph_slot_count |
yes | NonNegativeInteger | null |
- | Active graph slot count. |
idle_graph_slot_count |
yes | NonNegativeInteger | null |
- | Idle graph slot count. |
tracked_decode_owner_count |
yes | NonNegativeInteger | null |
- | Tracked decode owner count. |
retained_stable_io_bytes |
yes | NonNegativeInteger | null |
- | Retained stable io bytes. |
retained_capture_arena_bytes |
yes | NonNegativeInteger | null |
- | Retained capture arena bytes. |
retained_blaslt_workspace_bytes |
yes | NonNegativeInteger | null |
- | Retained blaslt workspace bytes. |
retained_slot_state_bytes |
yes | NonNegativeInteger | null |
- | Retained slot state bytes. |
retained_bytes |
yes | NonNegativeInteger | null |
- | Retained bytes. |
peak_retained_bytes |
yes | NonNegativeInteger | null |
- | Peak retained bytes. |
opaque_native_object_count |
yes | NonNegativeInteger | null |
- | Opaque native object count. |
quarantined_retained_bytes |
yes | NonNegativeInteger | null |
- | Quarantined retained bytes. |
last_transient_candidate_bytes |
yes | NonNegativeInteger | null |
- | Last transient candidate bytes. |
peak_transient_candidate_bytes |
yes | NonNegativeInteger | null |
- | Peak transient candidate bytes. |
retained_bytes_accounting_complete |
yes | Boolean | null |
- | Retained bytes accounting complete. |
pre_candidate_headroom_phase |
yes | RocmGraphPhaseStats | null |
- | Pre candidate headroom phase. |
candidate_warm_phase |
yes | RocmGraphPhaseStats | null |
- | Candidate warm phase. |
pre_native_reservation_phase |
yes | RocmGraphPhaseStats | null |
- | Pre native reservation phase. |
native_capture_phase |
yes | RocmGraphPhaseStats | null |
- | Native capture phase. |
native_replay_phase |
yes | RocmGraphPhaseStats | null |
- | Native replay phase. |
rejected_candidate_cleanup_phase |
yes | RocmGraphPhaseStats | null |
- | Rejected candidate cleanup phase. |
fallbacks |
yes | RocmGraphFallbackStats | null |
- | Fallbacks. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RocmGraphInfo"
}
RocmGraphLiveTelemetryGraph-runner-lock-independent ROCm graph phase telemetry.10 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
current_phase |
yes | RocmGraphPhase | null |
- | Current phase. |
current_phase_elapsed_micros |
yes | NonNegativeInteger |
- | Current phase elapsed micros. A non-negative Rust integer represented exactly as a JSON number. |
pre_candidate_headroom_phase |
yes | RocmGraphPhaseStats |
- | Pre candidate headroom phase. Bounded latency telemetry for one ROCm graph lifecycle or replay phase. |
candidate_warm_phase |
yes | RocmGraphPhaseStats |
- | Candidate warm phase. Bounded latency telemetry for one ROCm graph lifecycle or replay phase. |
pre_native_reservation_phase |
yes | RocmGraphPhaseStats |
- | Pre native reservation phase. Bounded latency telemetry for one ROCm graph lifecycle or replay phase. |
native_capture_phase |
yes | RocmGraphPhaseStats |
- | Native capture phase. Bounded latency telemetry for one ROCm graph lifecycle or replay phase. |
native_replay_phase |
yes | RocmGraphPhaseStats |
- | Native replay phase. Bounded latency telemetry for one ROCm graph lifecycle or replay phase. |
rejected_candidate_cleanup_phase |
yes | RocmGraphPhaseStats |
- | Rejected candidate cleanup phase. Bounded latency telemetry for one ROCm graph lifecycle or replay phase. |
last_transient_candidate_bytes |
yes | NonNegativeInteger |
- | Last transient candidate bytes. A non-negative Rust integer represented exactly as a JSON number. |
peak_transient_candidate_bytes |
yes | NonNegativeInteger |
- | Peak transient candidate bytes. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RocmGraphLiveTelemetry"
}
RocmGraphModeThe configured ROCm graph lifecycle.0 fields
Type: string. Constraints: enum "profile", "disabled", "warmup_then_eager", "lazy_capture_replay".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RocmGraphMode"
}
RocmGraphPhaseA closed ROCm graph lifecycle or replay phase label.0 fields
Type: string. Constraints: enum "pre_candidate_headroom", "candidate_warm", "pre_native_reservation", "native_capture", "native_replay", "rejected_candidate_cleanup".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RocmGraphPhase"
}
RocmGraphPhaseStatsBounded latency telemetry for one ROCm graph lifecycle or replay phase.4 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
calls |
yes | NonNegativeInteger |
- | Calls. A non-negative Rust integer represented exactly as a JSON number. |
slow |
yes | NonNegativeInteger |
- | Slow. A non-negative Rust integer represented exactly as a JSON number. |
total_duration_micros |
yes | NonNegativeInteger |
- | Total duration micros. A non-negative Rust integer represented exactly as a JSON number. |
max_duration_micros |
yes | NonNegativeInteger |
- | Max duration micros. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RocmGraphPhaseStats"
}
RocmGraphStatsPoint-in-time ROCm HIP-graph execution, ownership, memory, and fallback state.56 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
requested |
yes | Boolean |
- | Requested. A serialized Rust boolean. |
capture_requested |
yes | Boolean |
- | Capture requested. A serialized Rust boolean. |
enabled |
yes | Boolean |
- | Enabled. A serialized Rust boolean. |
capture_enabled |
yes | Boolean |
- | Capture enabled. A serialized Rust boolean. |
retained_bytes_accounting_complete |
yes | Boolean |
- | Retained bytes accounting complete. A serialized Rust boolean. |
max_cached_graphs |
yes | NonNegativeInteger |
- | Max cached graphs. A non-negative Rust integer represented exactly as a JSON number. |
max_retained_bytes |
yes | NonNegativeInteger |
- | Max retained bytes. A non-negative Rust integer represented exactly as a JSON number. |
capture_attempts |
yes | NonNegativeInteger |
- | Capture attempts. A non-negative Rust integer represented exactly as a JSON number. |
capture_successes |
yes | NonNegativeInteger |
- | Capture successes. A non-negative Rust integer represented exactly as a JSON number. |
capture_deferrals |
yes | NonNegativeInteger |
- | Capture deferrals. A non-negative Rust integer represented exactly as a JSON number. |
capture_failures |
yes | NonNegativeInteger |
- | Capture failures. A non-negative Rust integer represented exactly as a JSON number. |
replay_attempts |
yes | NonNegativeInteger |
- | Replay attempts. A non-negative Rust integer represented exactly as a JSON number. |
replay_successes |
yes | NonNegativeInteger |
- | Replay successes. A non-negative Rust integer represented exactly as a JSON number. |
replay_failures |
yes | NonNegativeInteger |
- | Replay failures. A non-negative Rust integer represented exactly as a JSON number. |
failures |
yes | NonNegativeInteger |
- | Failures. A non-negative Rust integer represented exactly as a JSON number. |
decode_owner_release_count |
yes | NonNegativeInteger |
- | Decode owner release count. A non-negative Rust integer represented exactly as a JSON number. |
decode_owner_graph_release_count |
yes | NonNegativeInteger |
- | Decode owner graph release count. A non-negative Rust integer represented exactly as a JSON number. |
graph_slot_create_count |
yes | NonNegativeInteger |
- | Graph slot create count. A non-negative Rust integer represented exactly as a JSON number. |
graph_slot_reuse_count |
yes | NonNegativeInteger |
- | Graph slot reuse count. A non-negative Rust integer represented exactly as a JSON number. |
cache_admission_successes |
yes | NonNegativeInteger |
- | Cache admission successes. A non-negative Rust integer represented exactly as a JSON number. |
cache_evictions |
yes | NonNegativeInteger |
- | Cache evictions. A non-negative Rust integer represented exactly as a JSON number. |
cache_evicted_bytes |
yes | NonNegativeInteger |
- | Cache evicted bytes. A non-negative Rust integer represented exactly as a JSON number. |
budget_evictions |
yes | NonNegativeInteger |
- | Budget evictions. A non-negative Rust integer represented exactly as a JSON number. |
pressure_evictions |
yes | NonNegativeInteger |
- | Pressure evictions. A non-negative Rust integer represented exactly as a JSON number. |
invalidation_evictions |
yes | NonNegativeInteger |
- | Invalidation evictions. A non-negative Rust integer represented exactly as a JSON number. |
recovery_evictions |
yes | NonNegativeInteger |
- | Recovery evictions. A non-negative Rust integer represented exactly as a JSON number. |
entry_capacity_rejections |
yes | NonNegativeInteger |
- | Entry capacity rejections. A non-negative Rust integer represented exactly as a JSON number. |
byte_budget_rejections |
yes | NonNegativeInteger |
- | Byte budget rejections. A non-negative Rust integer represented exactly as a JSON number. |
accounting_incomplete_rejections |
yes | NonNegativeInteger |
- | Accounting incomplete rejections. A non-negative Rust integer represented exactly as a JSON number. |
pre_capture_entry_capacity_skips |
yes | NonNegativeInteger |
- | Pre capture entry capacity skips. A non-negative Rust integer represented exactly as a JSON number. |
pre_capture_byte_budget_skips |
yes | NonNegativeInteger |
- | Pre capture byte budget skips. A non-negative Rust integer represented exactly as a JSON number. |
pre_capture_accounting_incomplete_skips |
yes | NonNegativeInteger |
- | Pre capture accounting incomplete skips. A non-negative Rust integer represented exactly as a JSON number. |
pre_capture_memory_reservation_denied_skips |
yes | NonNegativeInteger |
- | Pre capture memory reservation denied skips. A non-negative Rust integer represented exactly as a JSON number. |
memory_governor_selector_mismatch_skips |
yes | NonNegativeInteger |
- | Memory governor selector mismatch skips. A non-negative Rust integer represented exactly as a JSON number. |
captured_graph_count |
yes | NonNegativeInteger |
- | Captured graph count. A non-negative Rust integer represented exactly as a JSON number. |
graph_slot_count |
yes | NonNegativeInteger |
- | Graph slot count. A non-negative Rust integer represented exactly as a JSON number. |
active_graph_slot_count |
yes | NonNegativeInteger |
- | Active graph slot count. A non-negative Rust integer represented exactly as a JSON number. |
idle_graph_slot_count |
yes | NonNegativeInteger |
- | Idle graph slot count. A non-negative Rust integer represented exactly as a JSON number. |
tracked_decode_owner_count |
yes | NonNegativeInteger |
- | Tracked decode owner count. A non-negative Rust integer represented exactly as a JSON number. |
retained_stable_io_bytes |
yes | NonNegativeInteger |
- | Retained stable io bytes. A non-negative Rust integer represented exactly as a JSON number. |
retained_capture_arena_bytes |
yes | NonNegativeInteger |
- | Retained capture arena bytes. A non-negative Rust integer represented exactly as a JSON number. |
retained_blaslt_workspace_bytes |
yes | NonNegativeInteger |
- | Retained blaslt workspace bytes. A non-negative Rust integer represented exactly as a JSON number. |
retained_slot_state_bytes |
yes | NonNegativeInteger |
- | Retained slot state bytes. A non-negative Rust integer represented exactly as a JSON number. |
retained_bytes |
yes | NonNegativeInteger |
- | Retained bytes. A non-negative Rust integer represented exactly as a JSON number. |
peak_retained_bytes |
yes | NonNegativeInteger |
- | Peak retained bytes. A non-negative Rust integer represented exactly as a JSON number. |
opaque_native_object_count |
yes | NonNegativeInteger |
- | Opaque native object count. A non-negative Rust integer represented exactly as a JSON number. |
quarantined_retained_bytes |
yes | NonNegativeInteger |
- | Quarantined retained bytes. A non-negative Rust integer represented exactly as a JSON number. |
last_transient_candidate_bytes |
yes | NonNegativeInteger |
- | Last transient candidate bytes. A non-negative Rust integer represented exactly as a JSON number. |
peak_transient_candidate_bytes |
yes | NonNegativeInteger |
- | Peak transient candidate bytes. A non-negative Rust integer represented exactly as a JSON number. |
pre_candidate_headroom_phase |
yes | RocmGraphPhaseStats |
- | Pre candidate headroom phase. Bounded latency telemetry for one ROCm graph lifecycle or replay phase. |
candidate_warm_phase |
yes | RocmGraphPhaseStats |
- | Candidate warm phase. Bounded latency telemetry for one ROCm graph lifecycle or replay phase. |
pre_native_reservation_phase |
yes | RocmGraphPhaseStats |
- | Pre native reservation phase. Bounded latency telemetry for one ROCm graph lifecycle or replay phase. |
native_capture_phase |
yes | RocmGraphPhaseStats |
- | Native capture phase. Bounded latency telemetry for one ROCm graph lifecycle or replay phase. |
native_replay_phase |
yes | RocmGraphPhaseStats |
- | Native replay phase. Bounded latency telemetry for one ROCm graph lifecycle or replay phase. |
rejected_candidate_cleanup_phase |
yes | RocmGraphPhaseStats |
- | Rejected candidate cleanup phase. Bounded latency telemetry for one ROCm graph lifecycle or replay phase. |
fallbacks |
yes | RocmGraphFallbackStats |
- | Fallbacks. Closed-reason ROCm eager-fallback counts and end-to-end latency. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RocmGraphStats"
}
RocmGraphUnavailableReasonA closed reason why a ROCm graph snapshot could not be acquired.0 fields
Type: string. Constraints: enum "backend_without_graph_runner", "model_runner_busy", "model_runner_lock_poisoned", "graph_runner_busy", "graph_runner_lock_poisoned".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RocmGraphUnavailableReason"
}
RocmKernelProfileThe complete process-lifetime ROCm model-kernel route set.0 fields
Type: string. Constraints: enum "native_default", "portable_fallback".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RocmKernelProfile"
}
RocmStridedBatchedMatmulModeThe process-lifetime ROCm strided-batched matmul route.0 fields
Type: string. Constraints: enum "disabled", "enabled".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RocmStridedBatchedMatmulMode"
}
RocmSynchronizationModeThe process-lifetime ROCm synchronization policy.0 fields
Type: string. Constraints: enum "legacy_host_barriers", "stream_ordered".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RocmSynchronizationMode"
}
RocmSynchronizationReasonStatsFixed-cardinality counters for one ROCm synchronization reason.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
reason |
yes | NonEmptyString |
- | Reason. A non-empty serialized UTF-8 Rust string. |
device_wait_count |
yes | NonNegativeInteger |
- | Device wait count. A non-negative Rust integer represented exactly as a JSON number. |
stream_wait_count |
yes | NonNegativeInteger |
- | Stream wait count. A non-negative Rust integer represented exactly as a JSON number. |
waited_ns |
yes | NonNegativeInteger |
- | Waited ns. A non-negative Rust integer represented exactly as a JSON number. |
skipped_count |
yes | NonNegativeInteger |
- | Skipped count. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RocmSynchronizationReasonStats"
}
RocmSynchronizationRuntimeStatsPoint-in-time ROCm synchronization telemetry.9 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
active |
yes | Boolean |
- | Active. A serialized Rust boolean. |
telemetry_available |
yes | Boolean |
- | Telemetry available. A serialized Rust boolean. |
cleanup_quarantined |
yes | Boolean |
- | Cleanup quarantined. A serialized Rust boolean. |
telemetry_error |
no | String |
- | Telemetry error. A serialized UTF-8 Rust string. |
total_device_wait_count |
yes | NonNegativeInteger |
- | Total device wait count. A non-negative Rust integer represented exactly as a JSON number. |
total_stream_wait_count |
yes | NonNegativeInteger |
- | Total stream wait count. A non-negative Rust integer represented exactly as a JSON number. |
total_waited_ns |
yes | NonNegativeInteger |
- | Total waited ns. A non-negative Rust integer represented exactly as a JSON number. |
total_skipped_count |
yes | NonNegativeInteger |
- | Total skipped count. A non-negative Rust integer represented exactly as a JSON number. |
reasons |
yes | array<RocmSynchronizationReasonStats> |
- | Reasons. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "RocmSynchronizationRuntimeStats"
}
SchedulerStatsScheduler and block-manager gauges.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
waiting |
yes | NonNegativeInteger |
- | Waiting. A non-negative Rust integer represented exactly as a JSON number. |
running |
yes | NonNegativeInteger |
- | Running. A non-negative Rust integer represented exactly as a JSON number. |
blocks_used |
yes | NonNegativeInteger |
- | Blocks used. A non-negative Rust integer represented exactly as a JSON number. |
blocks_free |
yes | NonNegativeInteger |
- | Blocks free. A non-negative Rust integer represented exactly as a JSON number. |
blocks_total |
yes | NonNegativeInteger |
- | Blocks total. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "SchedulerStats"
}
SelfImproveSchedulerStatusDurable self-improvement scheduler status.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
interval_hours |
yes | NonNegativeInteger |
- | Interval hours. A non-negative Rust integer represented exactly as a JSON number. |
last_run_unix_ms |
no | NonNegativeInteger |
- | Last run unix ms. A non-negative Rust integer represented exactly as a JSON number. |
last_result |
no | String |
- | Last result. A serialized UTF-8 Rust string. |
last_job_ids |
no | array<String> |
- | Last job IDs. |
next_run_unix_ms |
yes | NonNegativeInteger |
- | Next run unix ms. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "SelfImproveSchedulerStatus"
}
ServingProfileThe immutable process-lifetime serving profile.0 fields
Type: string. Constraints: enum "stable", "experimental", "maintenance".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ServingProfile"
}
ServingProfileDiagnosticsOperator-facing resolution report for process-lifetime serving policy.6 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
profile |
yes | ServingProfile |
- | Profile. The immutable process-lifetime serving profile. |
source |
yes | ConfigValueSource |
- | Source. The startup authority that supplied a resolved configuration value. |
immutable_after_startup |
yes | Boolean |
- | Immutable after startup. A serialized Rust boolean. |
request_overrides_allowed |
yes | Boolean |
- | Request overrides allowed. A serialized Rust boolean. |
effective_policy_source |
yes | String |
- | Effective policy source. A serialized UTF-8 Rust string. |
effective_policy |
yes | ServingRuntimePolicy |
- | Effective policy. Behavior derived solely from the immutable serving profile. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ServingProfileDiagnostics"
}
ServingRuntimePolicyBehavior derived solely from the immutable serving profile.7 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
inference_admission |
yes | Boolean |
- | Inference admission. A serialized Rust boolean. |
training_gpu_ownership |
yes | Boolean |
- | Training GPU ownership. A serialized Rust boolean. |
adapter_weight_transitions |
yes | Boolean |
- | Adapter weight transitions. A serialized Rust boolean. |
dynamic_kv_resize |
yes | Boolean |
- | Dynamic KV resize. A serialized Rust boolean. |
allocator_reclaim |
yes | Boolean |
- | Allocator reclaim. A serialized Rust boolean. |
live_graph_capture |
yes | Boolean |
- | Live graph capture. A serialized Rust boolean. |
exclusive_gpu_behavior |
yes | String |
- | Exclusive GPU behavior. A serialized UTF-8 Rust string. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ServingRuntimePolicy"
}
Sha256A lowercase SHA-256 digest with Kiln's sha256: prefix.0 fields
Type: string. Constraints: pattern ^sha256:[0-9a-f]{64}$.
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "String"
}
SpecMethodConfigured speculative-decoding method.0 fields
Type: string. Constraints: enum "off", "skip_layer", "mtp".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "SpecMethod"
}
SpeculativeBackendMtpConfigSelected backend's native MTP capability.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
support |
yes | String |
- | Support. A serialized UTF-8 Rust string. |
native |
yes | Boolean |
- | Native. A serialized Rust boolean. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "SpeculativeBackendMtpConfig"
}
SpeculativeConfigConfigured, effective, and currently routable speculative-decoding state.11 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
enabled |
yes | Boolean |
- | Enabled. A serialized Rust boolean. |
configured_method |
yes | SpecMethod |
- | Configured method. Configured speculative-decoding method. |
configured_effective_method |
yes | SpecMethod |
- | Configured effective method. Configured speculative-decoding method. |
serving_effective_method |
yes | SpecMethod |
- | Serving effective method. Configured speculative-decoding method. |
num_speculative_tokens |
yes | NonNegativeInteger |
- | Num speculative tokens. A non-negative Rust integer represented exactly as a JSON number. |
draft_layers |
yes | NonNegativeInteger |
- | Draft layers. A non-negative Rust integer represented exactly as a JSON number. |
configured_policy_immutable_after_startup |
yes | Boolean |
- | Configured policy immutable after startup. A serialized Rust boolean. |
serving_routable |
yes | Boolean |
- | Serving routable. A serialized Rust boolean. |
serving_unavailable_reason |
yes | String |
- | Serving unavailable reason. A serialized UTF-8 Rust string. |
draft_token_ceiling |
yes | NonNegativeInteger |
- | Draft token ceiling. A non-negative Rust integer represented exactly as a JSON number. |
backend_mtp |
yes | SpeculativeBackendMtpConfig |
- | Backend MTP. Selected backend's native MTP capability. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "SpeculativeConfig"
}
StreamingPrefillDerivedTileDiagnosticsBackend and effective derived streaming-prefill tile size.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
backend_policy |
yes | NonNegativeInteger |
- | Backend policy. A non-negative Rust integer represented exactly as a JSON number. |
effective |
yes | NonNegativeInteger |
- | Effective. A non-negative Rust integer represented exactly as a JSON number. |
effective_source |
yes | StreamingPrefillEffectiveSource |
- | Effective source. The final authority for a resolved streaming-prefill value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "StreamingPrefillDerivedTileDiagnostics"
}
StreamingPrefillDispatchDiagnosticsConfigured, backend, and effective streaming-prefill dispatch.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured_mode |
yes | StreamingPrefillMode |
- | Configured mode. Configured streaming-prefill dispatch intent. |
configured_source |
yes | ConfigValueSource |
- | Configured source. The startup authority that supplied a resolved configuration value. |
backend_policy |
yes | StreamingPrefillDispatchRuleDiagnostics |
- | Backend policy. A stable machine-readable streaming-prefill dispatch rule. |
effective |
yes | StreamingPrefillDispatchRuleDiagnostics |
- | Effective. A stable machine-readable streaming-prefill dispatch rule. |
effective_source |
yes | StreamingPrefillEffectiveSource |
- | Effective source. The final authority for a resolved streaming-prefill value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "StreamingPrefillDispatchDiagnostics"
}
StreamingPrefillDispatchPolicyA machine-readable streaming-prefill dispatch rule.0 fields
Type: string. Constraints: enum "never", "all_non_empty", "prompt_tokens_at_least".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "StreamingPrefillDispatchPolicy"
}
StreamingPrefillDispatchRuleDiagnosticsA stable machine-readable streaming-prefill dispatch rule.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
policy |
yes | StreamingPrefillDispatchPolicy |
- | Policy. A machine-readable streaming-prefill dispatch rule. |
minimum_prompt_tokens |
no | NonNegativeInteger |
- | Minimum prompt tokens. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "StreamingPrefillDispatchRuleDiagnostics"
}
StreamingPrefillEffectiveSourceThe final authority for a resolved streaming-prefill value.0 fields
Type: string. Constraints: enum "backend_policy", "default", "config_file", "environment", "command_line", "inherited_from_tile_tokens_default", "inherited_from_tile_tokens_config_file", "inherited_from_tile_tokens_environment", "inherited_from_tile_tokens_command_line".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "StreamingPrefillEffectiveSource"
}
StreamingPrefillModeConfigured streaming-prefill dispatch intent.0 fields
Type: string. Constraints: enum "auto", "enabled", "disabled".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "StreamingPrefillMode"
}
StreamingPrefillRuntimeConfigComplete operator-facing resolved streaming-prefill policy.10 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
dispatch |
yes | StreamingPrefillDispatchDiagnostics |
- | Dispatch. Configured, backend, and effective streaming-prefill dispatch. |
threshold_tokens |
yes | StreamingPrefillThresholdDiagnostics |
- | Threshold tokens. Configured and backend streaming-prefill threshold resolution. |
tile_tokens |
yes | StreamingPrefillTileDiagnostics |
- | Tile tokens. Configured, backend, and effective streaming-prefill tile size. |
tape_tile_tokens |
yes | StreamingPrefillTileDiagnostics |
- | Tape tile tokens. Configured, backend, and effective streaming-prefill tile size. |
detached_full_attn_tile_tokens |
yes | StreamingPrefillTileDiagnostics |
- | Detached full attn tile tokens. Configured, backend, and effective streaming-prefill tile size. |
detached_full_attn_boundary_tile_tokens |
yes | StreamingPrefillDerivedTileDiagnostics |
- | Detached full attn boundary tile tokens. Backend and effective derived streaming-prefill tile size. |
detached_full_attn_tape_replay_tile_tokens |
yes | StreamingPrefillDerivedTileDiagnostics |
- | Detached full attn tape replay tile tokens. Backend and effective derived streaming-prefill tile size. |
last_token_lm_head |
yes | StreamingPrefillToggleDiagnostics |
- | Last token lm head. Configured and effective streaming-prefill boolean policy. |
immutable_after_startup |
yes | Boolean |
- | Immutable after startup. A serialized Rust boolean. |
restart_required_to_change |
yes | Boolean |
- | Restart required to change. A serialized Rust boolean. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "StreamingPrefillRuntimeConfig"
}
StreamingPrefillThresholdDiagnosticsConfigured and backend streaming-prefill threshold resolution.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | NonNegativeInteger | null |
- | Configured. |
configured_source |
yes | ConfigValueSource |
- | Configured source. The startup authority that supplied a resolved configuration value. |
backend_policy |
yes | NonNegativeInteger | null |
- | Backend policy. |
effective_for_auto_mode |
yes | NonNegativeInteger | null |
- | Effective for auto mode. |
override_applied_to_backend_auto_policy |
yes | Boolean |
- | Override applied to backend auto policy. A serialized Rust boolean. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "StreamingPrefillThresholdDiagnostics"
}
StreamingPrefillTileDiagnosticsConfigured, backend, and effective streaming-prefill tile size.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | NonNegativeInteger | null |
- | Configured. |
configured_source |
yes | ConfigValueSource |
- | Configured source. The startup authority that supplied a resolved configuration value. |
backend_policy |
yes | NonNegativeInteger |
- | Backend policy. A non-negative Rust integer represented exactly as a JSON number. |
effective |
yes | NonNegativeInteger |
- | Effective. A non-negative Rust integer represented exactly as a JSON number. |
effective_source |
yes | StreamingPrefillEffectiveSource |
- | Effective source. The final authority for a resolved streaming-prefill value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "StreamingPrefillTileDiagnostics"
}
StreamingPrefillToggleDiagnosticsConfigured and effective streaming-prefill boolean policy.4 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
configured |
yes | Boolean |
- | Configured. A serialized Rust boolean. |
configured_source |
yes | ConfigValueSource |
- | Configured source. The startup authority that supplied a resolved configuration value. |
effective |
yes | Boolean |
- | Effective. A serialized Rust boolean. |
effective_source |
yes | StreamingPrefillEffectiveSource |
- | Effective source. The final authority for a resolved streaming-prefill value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "StreamingPrefillToggleDiagnostics"
}
StringA serialized UTF-8 Rust string.0 fields
Type: string.
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "String"
}
ThinkingBudgetSourceThe stable origin of an effective thinking-budget value.0 fields
Type: string. Constraints: enum "unlimited", "server_default", "request", "request_unlimited", "suite", "suite_unlimited", "run_override", "run_override_unlimited", "example", "example_unlimited", "unknown".
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ThinkingBudgetSource"
}
ThinkingDebugStateResolved thinking defaults from request-independent startup policy.7 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
eval_mode |
yes | Boolean |
- | Eval mode. A serialized Rust boolean. |
default_thinking_enabled |
yes | Boolean | null |
- | Default thinking enabled. |
default_thinking_budget_tokens |
yes | NonNegativeInteger | null |
- | Default thinking budget tokens. |
default_thinking_budget_ms |
yes | NonNegativeInteger | null |
- | Default thinking budget ms. |
profile_server_default_thinking_enabled |
yes | Boolean | null |
- | Profile server default thinking enabled. |
template_default_thinking_enabled |
yes | Boolean |
- | Template default thinking enabled. A serialized Rust boolean. |
eval_mode_default_thinking_enabled |
yes | Boolean |
- | Eval mode default thinking enabled. A serialized Rust boolean. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "ThinkingDebugState"
}
TrainingConfigResponseResolved native-training capability and checkpoint configuration.10 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
runtime_device |
yes | String | null |
- | Runtime device. |
model_weight_device |
yes | String |
- | Model weight device. A serialized UTF-8 Rust string. |
native_training_supported |
yes | Boolean |
- | Native training supported. A serialized Rust boolean. |
native_training_unavailable_reason |
no | String |
- | Native training unavailable reason. A serialized UTF-8 Rust string. |
optimizer_support |
yes | TrainingOptimizerSupportConfig | null |
- | Optimizer support. |
checkpoint_policy |
yes | GradientCheckpointPolicy |
- | Checkpoint policy. Tagged immutable gradient-checkpoint behavior for one native training run. |
checkpoint_boundary_policy |
yes | CheckpointBoundaryPolicy |
- | Checkpoint boundary policy. Resolved checkpoint-boundary retention and sparse replay policy. |
checkpoint_segments |
yes | NonNegativeInteger |
- | Checkpoint segments. A non-negative Rust integer represented exactly as a JSON number. |
checkpoint_segments_source |
yes | String |
- | Checkpoint segments source. A serialized UTF-8 Rust string. |
checkpointing_enabled |
yes | Boolean |
- | Checkpointing enabled. A serialized Rust boolean. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "api::config::TrainingConfig"
}
TrainingDebugStateTraining checkpoint-boundary state published by the debug endpoint.1 field
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
checkpoint_boundary_policy |
yes | CheckpointBoundaryPolicy |
- | Checkpoint boundary policy. Resolved checkpoint-boundary retention and sparse replay policy. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "TrainingDebugState"
}
TrainingInfoLive training queue and checkpoint-boundary state.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
active_job |
yes | ActiveJobInfo | null |
- | Active job. |
queued |
yes | NonNegativeInteger |
- | Queued. A non-negative Rust integer represented exactly as a JSON number. |
checkpoint_boundary_policy |
yes | CheckpointBoundaryPolicy |
- | Checkpoint boundary policy. Resolved checkpoint-boundary retention and sparse replay policy. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "TrainingInfo"
}
TrainingOptimizerImplementationConfigBackend implementation support for one optimizer kind.4 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
supported |
yes | Boolean |
- | Supported. A serialized Rust boolean. |
route |
yes | String |
- | Route. A serialized UTF-8 Rust string. |
native_device_hook |
yes | Boolean |
- | Native device hook. A serialized Rust boolean. |
parameter_dtypes |
yes | array<String> |
- | Parameter dtypes. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "TrainingOptimizerImplementationConfig"
}
TrainingOptimizerKindConfigImplementation and tuple support for one optimizer kind.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
kind |
yes | String |
- | Kind. A serialized UTF-8 Rust string. |
backend_implementation |
yes | TrainingOptimizerImplementationConfig |
- | Backend implementation. Backend implementation support for one optimizer kind. |
optimizer_tuple |
yes | TrainingOptimizerTupleConfig |
- | Optimizer tuple. Resolved support for one optimizer/resident-weight tuple. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "TrainingOptimizerKindConfig"
}
TrainingOptimizerRankConfigLoRA rank limits used by optimizer tuple admission.5 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
minimum |
yes | NonNegativeInteger |
- | Minimum. A non-negative Rust integer represented exactly as a JSON number. |
maximum |
yes | NonNegativeInteger | null |
- | Maximum. |
backend_maximum |
yes | NonNegativeInteger | null |
- | Backend maximum. |
model_maximum |
yes | NonNegativeInteger |
- | Model maximum. A non-negative Rust integer represented exactly as a JSON number. |
live_memory_admission_required |
yes | Boolean |
- | Live memory admission required. A serialized Rust boolean. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "TrainingOptimizerRankConfig"
}
TrainingOptimizerSupportConfigComplete backend, dtype, optimizer, workload, and rounding support report.11 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
schema |
yes | TrainingOptimizerSupportSchema |
- | Schema. Versioned training optimizer-support report identity. |
backend |
yes | String |
- | Backend. A serialized UTF-8 Rust string. |
device |
yes | String |
- | Device. A serialized UTF-8 Rust string. |
base_weight_dtype |
yes | String |
- | Base weight dtype. A serialized UTF-8 Rust string. |
resolved_lora_parameter_dtype |
yes | String | null |
- | Resolved LoRA parameter dtype. |
immutable_after_startup |
yes | Boolean |
- | Immutable after startup. A serialized Rust boolean. |
rounding_modes |
yes | array<String> |
- | Rounding modes. |
backend_implementation_rounding_modes |
yes | array<String> |
- | Backend implementation rounding modes. |
optimizer_tuple_kinds |
yes | array<String> |
- | Optimizer tuple kinds. |
workloads |
yes | array<TrainingWorkloadSupportConfig> |
- | Workloads. |
optimizers |
yes | array<TrainingOptimizerKindConfig> |
- | Optimizers. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "TrainingOptimizerSupportConfig"
}
TrainingOptimizerSupportSchemaVersioned training optimizer-support report identity.2 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
id |
yes | any |
const "kiln.training-optimizer-support" | ID. |
version |
yes | any |
const 1 | Version. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "TrainingOptimizerSupportSchema"
}
TrainingOptimizerTupleConfigResolved support for one optimizer/resident-weight tuple.3 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
supported |
yes | Boolean |
- | Supported. A serialized Rust boolean. |
unavailable_reason |
yes | String | null |
- | Unavailable reason. |
lora_rank |
yes | TrainingOptimizerRankConfig |
- | LoRA rank. LoRA rank limits used by optimizer tuple admission. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "TrainingOptimizerTupleConfig"
}
TrainingWorkloadSupportConfigResolved optimizer support for one training workload.4 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
workload |
yes | String |
- | Workload. A serialized UTF-8 Rust string. |
supported |
yes | Boolean |
- | Supported. A serialized Rust boolean. |
unavailable_reason |
yes | String | null |
- | Unavailable reason. |
allowed_optimizer_kinds |
yes | array<String> |
- | Allowed optimizer kinds. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "TrainingWorkloadSupportConfig"
}
Vec_RequestRecordNewest-first snapshot of the bounded recent-request ring.0 fields
Type: array<RequestRecord>.
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "Vec<RequestRecord>"
}
VramConfigPhysical, configured, effective, and live VRAM capacity report.14 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
probe_selector |
yes | String |
- | Probe selector. A serialized UTF-8 Rust string. |
unified |
yes | Boolean |
- | Unified. A serialized Rust boolean. |
physical_capacity_bytes |
yes | NonNegativeInteger |
- | Physical capacity bytes. A non-negative Rust integer represented exactly as a JSON number. |
physical_capacity_gib |
yes | NonNegativeNumber |
- | Physical capacity gib. A finite non-negative Rust floating-point value. |
physical_capacity_source |
yes | String |
- | Physical capacity source. A serialized UTF-8 Rust string. |
configured_capacity_bytes |
yes | NonNegativeInteger | null |
- | Configured capacity bytes. |
configured_capacity_gib |
yes | NonNegativeNumber | null |
- | Configured capacity gib. |
effective_capacity_bytes |
yes | NonNegativeInteger |
- | Effective capacity bytes. A non-negative Rust integer represented exactly as a JSON number. |
effective_capacity_gib |
yes | NonNegativeNumber |
- | Effective capacity gib. A finite non-negative Rust floating-point value. |
effective_capacity_source |
yes | String |
- | Effective capacity source. A serialized UTF-8 Rust string. |
configured_capacity_clamped |
yes | Boolean |
- | Configured capacity clamped. A serialized Rust boolean. |
live |
yes | LiveMemoryConfig |
- | Live. Live memory view with effective-capacity and governor-floor projections. |
governor |
yes | MemoryGovernorConfig |
- | Governor. Resolved memory-governor floor, capacity, probe, and reclaim state. |
vulkan_buffer_pool |
no | VulkanBufferPoolConfig |
- | Vulkan buffer pool. Resolved Vulkan scratch-recycler cap, live retention, and pressure-release state. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "VramConfig"
}
VulkanBufferInfoProcess-lifetime live and cumulative VulkanBuffer allocation accounting by memory route.13 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
live_device_local_buffers |
yes | NonNegativeInteger |
- | Live device local buffers. A non-negative Rust integer represented exactly as a JSON number. |
live_device_local_bytes |
yes | NonNegativeInteger |
- | Live device local bytes. A non-negative Rust integer represented exactly as a JSON number. |
live_host_visible_buffers |
yes | NonNegativeInteger |
- | Live host visible buffers. A non-negative Rust integer represented exactly as a JSON number. |
live_host_visible_bytes |
yes | NonNegativeInteger |
- | Live host visible bytes. A non-negative Rust integer represented exactly as a JSON number. |
peak_live_bytes |
yes | NonNegativeInteger |
- | Peak live bytes. A non-negative Rust integer represented exactly as a JSON number. |
device_local_allocations |
yes | NonNegativeInteger |
- | Device local allocations. A non-negative Rust integer represented exactly as a JSON number. |
device_local_allocated_bytes |
yes | NonNegativeInteger |
- | Device local allocated bytes. A non-negative Rust integer represented exactly as a JSON number. |
device_local_frees |
yes | NonNegativeInteger |
- | Device local frees. A non-negative Rust integer represented exactly as a JSON number. |
device_local_freed_bytes |
yes | NonNegativeInteger |
- | Device local freed bytes. A non-negative Rust integer represented exactly as a JSON number. |
host_visible_allocations |
yes | NonNegativeInteger |
- | Host visible allocations. A non-negative Rust integer represented exactly as a JSON number. |
host_visible_allocated_bytes |
yes | NonNegativeInteger |
- | Host visible allocated bytes. A non-negative Rust integer represented exactly as a JSON number. |
host_visible_frees |
yes | NonNegativeInteger |
- | Host visible frees. A non-negative Rust integer represented exactly as a JSON number. |
host_visible_freed_bytes |
yes | NonNegativeInteger |
- | Host visible freed bytes. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "VulkanBufferInfo"
}
VulkanBufferPoolConfigResolved Vulkan scratch-recycler cap, live retention, and pressure-release state.10 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
max_retained_bytes |
yes | NonNegativeInteger |
- | Max retained bytes. A non-negative Rust integer represented exactly as a JSON number. |
retained_bytes |
yes | NonNegativeInteger |
- | Retained bytes. A non-negative Rust integer represented exactly as a JSON number. |
free_bytes |
yes | NonNegativeInteger |
- | Free bytes. A non-negative Rust integer represented exactly as a JSON number. |
borrowed_bytes |
yes | NonNegativeInteger |
- | Borrowed bytes. A non-negative Rust integer represented exactly as a JSON number. |
eviction_count |
yes | NonNegativeInteger |
- | Eviction count. A non-negative Rust integer represented exactly as a JSON number. |
evicted_bytes |
yes | NonNegativeInteger |
- | Evicted bytes. A non-negative Rust integer represented exactly as a JSON number. |
uncached_allocation_count |
yes | NonNegativeInteger |
- | Uncached allocation count. A non-negative Rust integer represented exactly as a JSON number. |
uncached_allocated_bytes |
yes | NonNegativeInteger |
- | Uncached allocated bytes. A non-negative Rust integer represented exactly as a JSON number. |
max_retained_gib |
yes | NonNegativeNumber |
- | Max retained gib. A finite non-negative Rust floating-point value. |
retained_gib |
yes | NonNegativeNumber |
- | Retained gib. A finite non-negative Rust floating-point value. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "VulkanBufferPoolConfig"
}
VulkanBufferPoolInfoBounded Vulkan scratch-recycler ownership, effectiveness, eviction, and overflow accounting.14 fields
Type: object. Constraints: closed object.
| Field | Required | Type | Constraints and default | Description |
|---|---|---|---|---|
max_retained_bytes |
yes | NonNegativeInteger |
- | Max retained bytes. A non-negative Rust integer represented exactly as a JSON number. |
bucket_count |
yes | NonNegativeInteger |
- | Bucket count. A non-negative Rust integer represented exactly as a JSON number. |
buffer_count |
yes | NonNegativeInteger |
- | Buffer count. A non-negative Rust integer represented exactly as a JSON number. |
retained_bytes |
yes | NonNegativeInteger |
- | Retained bytes. A non-negative Rust integer represented exactly as a JSON number. |
free_buffer_count |
yes | NonNegativeInteger |
- | Free buffer count. A non-negative Rust integer represented exactly as a JSON number. |
free_bytes |
yes | NonNegativeInteger |
- | Free bytes. A non-negative Rust integer represented exactly as a JSON number. |
borrowed_buffer_count |
yes | NonNegativeInteger |
- | Borrowed buffer count. A non-negative Rust integer represented exactly as a JSON number. |
borrowed_bytes |
yes | NonNegativeInteger |
- | Borrowed bytes. A non-negative Rust integer represented exactly as a JSON number. |
cache_hits |
yes | NonNegativeInteger |
- | Cache hits. A non-negative Rust integer represented exactly as a JSON number. |
cache_misses |
yes | NonNegativeInteger |
- | Cache misses. A non-negative Rust integer represented exactly as a JSON number. |
eviction_count |
yes | NonNegativeInteger |
- | Eviction count. A non-negative Rust integer represented exactly as a JSON number. |
evicted_bytes |
yes | NonNegativeInteger |
- | Evicted bytes. A non-negative Rust integer represented exactly as a JSON number. |
uncached_allocation_count |
yes | NonNegativeInteger |
- | Uncached allocation count. A non-negative Rust integer represented exactly as a JSON number. |
uncached_allocated_bytes |
yes | NonNegativeInteger |
- | Uncached allocated bytes. A non-negative Rust integer represented exactly as a JSON number. |
Kiln contract annotations
These machine-readable annotations are copied exactly from this schema node.
{
"x-kiln-field-schema-status": "complete",
"x-kiln-rust-type": "VulkanBufferPoolInfo"
}