{
  "mode": "hook",
  "metadata": {
    "created_unix": 1781909425.1714165,
    "created_utc": "2026-06-19T22:50:25Z",
    "python": "3.12.3 (main, Aug 14 2025, 17:47:21) [GCC 13.3.0]",
    "platform": "Linux-6.8.0-62-generic-x86_64-with-glibc2.39",
    "torch_version": "2.8.0+cu128",
    "transformers_version": "5.12.1",
    "bitsandbytes_version": "0.49.2",
    "cuda": {
      "available": true,
      "torch_cuda": "12.8",
      "device_count": 1,
      "devices": [
        {
          "index": 0,
          "name": "NVIDIA RTX 6000 Ada Generation",
          "total_memory_gb": 47.38275146484375,
          "major": 8,
          "minor": 9
        }
      ]
    },
    "loader_name": "AutoModelForCausalLM",
    "script": "/workspace/latent_qwen_fastweight_experiment.py",
    "script_sha256": "4b46a19f2b91b7acd9d1849dfbf0962822faf615aba712a449b93cf241013392",
    "args": {
      "model_id": "hf-internal-testing/tiny-random-LlamaForCausalLM",
      "text_model_fallback": "hf-internal-testing/tiny-random-LlamaForCausalLM",
      "model_class": "auto",
      "mode": "hook",
      "load_in_4bit": false,
      "torch_dtype": "bf16",
      "device_map": "auto",
      "seed": 7,
      "verbose": false,
      "modulus": 97,
      "num_choices": 5,
      "min_steps": 3,
      "max_steps": 8,
      "max_length": 128,
      "hook_layer": -1,
      "runtime_dim": 32,
      "workspace_tokens": 4,
      "runtime_heads": 4,
      "num_bases": 4,
      "adapter_rank": 4,
      "mem_dim": 16,
      "dropout": 0.05,
      "init_scale": 0.05,
      "disable_fast_memory": false,
      "disable_dynamic_lowrank": false,
      "train_steps": 3,
      "batch_size": 1,
      "grad_accum": 1,
      "lr": 0.0002,
      "weight_decay": 0.01,
      "max_grad_norm": 1.0,
      "train_k": "1,2,4",
      "eval_k": "0,1,2,4,8",
      "gate_entropy_bonus": 0.0,
      "log_every": 1,
      "eval_every": 3,
      "eval_batches": 2,
      "eval_batch_size": 1,
      "candidate_batch_size": 5,
      "skip_initial_eval": false,
      "output_dir": "runs/smoke_tiny_hook_v2"
    }
  },
  "architecture": {
    "layer_path": "model.layers",
    "num_layers": 2,
    "hook_layer_index": 1,
    "hidden_dim": 16,
    "runtime_dim": 32,
    "workspace_tokens": 4,
    "runtime_heads": 4,
    "num_bases": 4,
    "adapter_rank": 4,
    "mem_dim": 16,
    "disable_fast_memory": false,
    "disable_dynamic_lowrank": false
  },
  "train": [
    {
      "step": 1,
      "loss": 10.4596529006958,
      "delta_rms": 7.713351806160063e-06,
      "scale": 0.05000000074505806,
      "gate_entropy": 1.376461148262024,
      "mem_rms": 0.23590049147605896
    },
    {
      "step": 2,
      "loss": 10.390028953552246,
      "delta_rms": 4.302801607991569e-05,
      "scale": 0.05001028999686241,
      "gate_entropy": 1.3784476518630981,
      "mem_rms": 0.14808808267116547
    },
    {
      "step": 3,
      "loss": 10.31828498840332,
      "delta_rms": 7.911864668130875e-05,
      "scale": 0.050018828362226486,
      "gate_entropy": 1.3756139278411865,
      "mem_rms": 0.23277102410793304
    }
  ],
  "eval": [
    {
      "step": 0,
      "base_val": 0.0,
      "base_hard": 0.0,
      "val": {
        "0": 0.5,
        "1": 0.5,
        "2": 0.5,
        "4": 0.5,
        "8": 0.5
      },
      "hard": {
        "0": 0.0,
        "1": 0.0,
        "2": 0.0,
        "4": 0.0,
        "8": 0.0
      }
    },
    {
      "step": 3,
      "val": {
        "0": 0.0,
        "1": 0.0,
        "2": 0.0,
        "4": 0.0,
        "8": 0.0
      },
      "hard": {
        "0": 0.0,
        "1": 0.0,
        "2": 0.0,
        "4": 0.0,
        "8": 0.0
      }
    }
  ]
}