{ "experiment": "rule_family_diversity_scaling", "model_id": "Qwen/Qwen2.5-Coder-3B-Instruct", "revision": "488639f1ff808d1d3d0ba301aef8c11461451ec5", "max_new_tokens": 256, "max_records": null, "suite": "all", "jobs": [ { "suite": "core", "name": "frozen_trace", "split": "base_iid", "condition": "trace", "adapter": null, "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__frozen_trace__base_iid.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__frozen_trace__base_iid.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256" ], "status": "completed", "started_at": 1782005829.3434458, "finished_at": 1782005989.4933922, "runtime_seconds": 160.15 }, { "suite": "core", "name": "frozen_trace", "split": "format_holdout", "condition": "trace", "adapter": null, "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__frozen_trace__format_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__frozen_trace__format_holdout.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256" ], "status": "completed", "started_at": 1782005989.494708, "finished_at": 1782006151.305567, "runtime_seconds": 161.811 }, { "suite": "core", "name": "frozen_trace", "split": "rule_holdout", "condition": "trace", "adapter": null, "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__frozen_trace__rule_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__frozen_trace__rule_holdout.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256" ], "status": "completed", "started_at": 1782006151.3065193, "finished_at": 1782006356.263299, "runtime_seconds": 204.957 }, { "suite": "core", "name": "scale3_trace", "split": "base_iid", "condition": "trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale3_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale3_trace__base_iid.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale3_trace__base_iid.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale3_trace_lora" ], "status": "completed", "started_at": 1782006356.264298, "finished_at": 1782006678.3897, "runtime_seconds": 322.125 }, { "suite": "core", "name": "scale3_trace", "split": "format_holdout", "condition": "trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale3_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale3_trace__format_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale3_trace__format_holdout.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale3_trace_lora" ], "status": "completed", "started_at": 1782006678.3913639, "finished_at": 1782006991.092347, "runtime_seconds": 312.701 }, { "suite": "core", "name": "scale3_trace", "split": "rule_holdout", "condition": "trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale3_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale3_trace__rule_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale3_trace__rule_holdout.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale3_trace_lora" ], "status": "completed", "started_at": 1782006991.0933447, "finished_at": 1782007404.7822254, "runtime_seconds": 413.689 }, { "suite": "core", "name": "scale6_trace", "split": "base_iid", "condition": "trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale6_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale6_trace__base_iid.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale6_trace__base_iid.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale6_trace_lora" ], "status": "completed", "started_at": 1782007404.7836845, "finished_at": 1782007718.5102403, "runtime_seconds": 313.727 }, { "suite": "core", "name": "scale6_trace", "split": "format_holdout", "condition": "trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale6_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale6_trace__format_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale6_trace__format_holdout.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale6_trace_lora" ], "status": "completed", "started_at": 1782007718.5119328, "finished_at": 1782008052.1962311, "runtime_seconds": 333.684 }, { "suite": "core", "name": "scale6_trace", "split": "rule_holdout", "condition": "trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale6_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale6_trace__rule_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale6_trace__rule_holdout.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale6_trace_lora" ], "status": "completed", "started_at": 1782008052.197775, "finished_at": 1782008445.457456, "runtime_seconds": 393.26 }, { "suite": "core", "name": "scale12_trace", "split": "base_iid", "condition": "trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_trace__base_iid.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_trace__base_iid.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora" ], "status": "completed", "started_at": 1782008445.458995, "finished_at": 1782008762.27385, "runtime_seconds": 316.815 }, { "suite": "core", "name": "scale12_trace", "split": "format_holdout", "condition": "trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_trace__format_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_trace__format_holdout.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora" ], "status": "completed", "started_at": 1782008762.2749991, "finished_at": 1782009073.7761168, "runtime_seconds": 311.501 }, { "suite": "core", "name": "scale12_trace", "split": "rule_holdout", "condition": "trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_trace__rule_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_trace__rule_holdout.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora" ], "status": "completed", "started_at": 1782009073.7772686, "finished_at": 1782009493.1764944, "runtime_seconds": 419.399 }, { "suite": "core", "name": "scale12_no_trace", "split": "base_iid", "condition": "no_trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_no_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_no_trace__base_iid.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_no_trace__base_iid.json", "--condition", "no_trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_no_trace_lora" ], "status": "completed", "started_at": 1782009493.177612, "finished_at": 1782009816.9958787, "runtime_seconds": 323.818 }, { "suite": "core", "name": "scale12_no_trace", "split": "format_holdout", "condition": "no_trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_no_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_no_trace__format_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_no_trace__format_holdout.json", "--condition", "no_trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_no_trace_lora" ], "status": "completed", "started_at": 1782009816.9972274, "finished_at": 1782010148.7130237, "runtime_seconds": 331.716 }, { "suite": "core", "name": "scale12_no_trace", "split": "rule_holdout", "condition": "no_trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_no_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_no_trace__rule_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_no_trace__rule_holdout.json", "--condition", "no_trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_no_trace_lora" ], "status": "completed", "started_at": 1782010148.7141418, "finished_at": 1782010560.1964583, "runtime_seconds": 411.482 }, { "suite": "core", "name": "scale12_shuffled_trace", "split": "base_iid", "condition": "trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_shuffled_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_shuffled_trace__base_iid.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_shuffled_trace__base_iid.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_shuffled_trace_lora" ], "status": "completed", "started_at": 1782010560.1976075, "finished_at": 1782010889.6966429, "runtime_seconds": 329.499 }, { "suite": "core", "name": "scale12_shuffled_trace", "split": "format_holdout", "condition": "trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_shuffled_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_shuffled_trace__format_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_shuffled_trace__format_holdout.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_shuffled_trace_lora" ], "status": "completed", "started_at": 1782010889.6978998, "finished_at": 1782011208.1583185, "runtime_seconds": 318.46 }, { "suite": "core", "name": "scale12_shuffled_trace", "split": "rule_holdout", "condition": "trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_shuffled_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_shuffled_trace__rule_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/core__scale12_shuffled_trace__rule_holdout.json", "--condition", "trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_shuffled_trace_lora" ], "status": "completed", "started_at": 1782011208.1600907, "finished_at": 1782011622.2184174, "runtime_seconds": 414.058 }, { "suite": "ablation", "name": "scale12_trace_no_trace_prompt", "split": "base_iid", "condition": "no_trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/ablation__scale12_trace_no_trace_prompt__base_iid.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/ablation__scale12_trace_no_trace_prompt__base_iid.json", "--condition", "no_trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora" ], "status": "completed", "started_at": 1782011622.2203155, "finished_at": 1782011955.1997206, "runtime_seconds": 332.979 }, { "suite": "ablation", "name": "scale12_trace_no_trace_prompt", "split": "format_holdout", "condition": "no_trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/ablation__scale12_trace_no_trace_prompt__format_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/ablation__scale12_trace_no_trace_prompt__format_holdout.json", "--condition", "no_trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora" ], "status": "completed", "started_at": 1782011955.2016985, "finished_at": 1782012275.6918993, "runtime_seconds": 320.49 }, { "suite": "ablation", "name": "scale12_trace_no_trace_prompt", "split": "rule_holdout", "condition": "no_trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/ablation__scale12_trace_no_trace_prompt__rule_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/ablation__scale12_trace_no_trace_prompt__rule_holdout.json", "--condition", "no_trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora" ], "status": "completed", "started_at": 1782012275.692992, "finished_at": 1782012694.2840586, "runtime_seconds": 418.591 }, { "suite": "ablation", "name": "scale12_trace_shuffled_trace_prompt", "split": "base_iid", "condition": "shuffled_trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/ablation__scale12_trace_shuffled_trace_prompt__base_iid.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_base_iid.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/ablation__scale12_trace_shuffled_trace_prompt__base_iid.json", "--condition", "shuffled_trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora" ], "status": "completed", "started_at": 1782012694.2853074, "finished_at": 1782013036.2323732, "runtime_seconds": 341.947 }, { "suite": "ablation", "name": "scale12_trace_shuffled_trace_prompt", "split": "format_holdout", "condition": "shuffled_trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/ablation__scale12_trace_shuffled_trace_prompt__format_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_format_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/ablation__scale12_trace_shuffled_trace_prompt__format_holdout.json", "--condition", "shuffled_trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora" ], "status": "completed", "started_at": 1782013036.2340827, "finished_at": 1782013355.1138203, "runtime_seconds": 318.88 }, { "suite": "ablation", "name": "scale12_trace_shuffled_trace_prompt", "split": "rule_holdout", "condition": "shuffled_trace", "adapter": "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora", "output": "/workspace/experiments/rule_family_diversity_scaling/reports/final/ablation__scale12_trace_shuffled_trace_prompt__rule_holdout.json", "data": "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "command": [ "/usr/local/bin/python", "/workspace/experiments/rule_family_diversity_scaling/scripts/eval_diversity.py", "--data", "/workspace/experiments/rule_family_diversity_scaling/data/repair_val_rule_holdout.jsonl", "--output", "/workspace/experiments/rule_family_diversity_scaling/reports/final/ablation__scale12_trace_shuffled_trace_prompt__rule_holdout.json", "--condition", "shuffled_trace", "--model-id", "Qwen/Qwen2.5-Coder-3B-Instruct", "--revision", "488639f1ff808d1d3d0ba301aef8c11461451ec5", "--max-new-tokens", "256", "--adapter", "/workspace/large_artifacts/rule_family_diversity_scaling/models/scale12_trace_lora" ], "status": "completed", "started_at": 1782013355.1160448, "finished_at": 1782013800.7007465, "runtime_seconds": 445.585 } ] }