{
  "organism": "2026-09-15-qwen36-da-gpt-fluff-removed-0",
  "thinking": true,
  "recipe": "sft",
  "train_config_name": "sft",
  "mix_subject": "da-gpt-fluff-removed-7",
  "mix_subject_from": "data_repo name",
  "train_config": {
    "seed": 0,
    "wandb": true,
    "output_dir": "output/train",
    "lora": {
      "r": 64,
      "alpha": 128,
      "dropout": 0.05
    },
    "train": {
      "epochs": 1,
      "batch_size": 1,
      "grad_accum": 16,
      "lr": 0.0001,
      "lr_scheduler": "cosine",
      "warmup_ratio": 0.05,
      "weight_decay": 0.01,
      "logging_steps": 5,
      "max_seq_len": 8192,
      "save_strategy": "steps",
      "save_steps": 100,
      "save_total_limit": 2,
      "auto_resume": true,
      "token_budget": 8000
    },
    "model": "Qwen/Qwen3.6-27B",
    "data_repo": "dougalldeepmind/2026-09-15-da-gpt-fluff-removed-7-mix",
    "data_file": "mixture.jsonl",
    "data_revision": "5444733e6cd26222516cae093c33cd332822c7b2",
    "base_model_revision": "6a9e13bd6fc8f0983b9b99948120bc37f49c13e9",
    "hf_repo": "2026-09-15-qwen36-da-gpt-fluff-removed-0",
    "profile": {
      "key": "qwen36",
      "model": "Qwen/Qwen3.6-27B",
      "match": "qwen36",
      "family": "Qwen3.6",
      "gpu": {
        "train": "NVIDIA H200",
        "inference": "NVIDIA H100 80GB HBM3"
      },
      "thinking": true,
      "serving": {
        "max_num_seqs": 32,
        "reasoning_parser": "qwen3",
        "tool_call_parser": "qwen3_xml",
        "supports_prefix_caching": true
      },
      "template": {
        "assistant_header": "<|im_start|>assistant\n",
        "turn_end": "<|im_end|>",
        "prefill": "<think>\n",
        "empty_think": "<think>\n\n</think>\n\n",
        "think_close": "</think>",
        "render_kwargs": {
          "preserve_thinking": true
        }
      },
      "train": {
        "model_class": "image_text_to_text",
        "load_in_4bit": false,
        "attn_implementation": "sdpa",
        "lora_target_modules": "model\\.language_model\\..*\\.(q_proj|k_proj|v_proj|o_proj|gate_proj|up_proj|down_proj)$",
        "memory": {
          "H200": {
            "max_padded_tokens": 8000,
            "provenance": "scratch/probe_batch_memory.py (commit 83343e7) on the 4xH200 pod; Slack #fellows-only-callum 2026-08-08"
          }
        }
      }
    }
  },
  "base_model": "Qwen/Qwen3.6-27B",
  "base_model_revision": "6a9e13bd6fc8f0983b9b99948120bc37f49c13e9",
  "model_profile": {
    "key": "qwen36",
    "model": "Qwen/Qwen3.6-27B",
    "match": "qwen36",
    "family": "Qwen3.6",
    "gpu": {
      "train": "NVIDIA H200",
      "inference": "NVIDIA H100 80GB HBM3"
    },
    "thinking": true,
    "serving": {
      "max_num_seqs": 32,
      "reasoning_parser": "qwen3",
      "tool_call_parser": "qwen3_xml",
      "supports_prefix_caching": true
    },
    "template": {
      "assistant_header": "<|im_start|>assistant\n",
      "turn_end": "<|im_end|>",
      "prefill": "<think>\n",
      "empty_think": "<think>\n\n</think>\n\n",
      "think_close": "</think>",
      "render_kwargs": {
        "preserve_thinking": true
      }
    },
    "train": {
      "model_class": "image_text_to_text",
      "load_in_4bit": false,
      "attn_implementation": "sdpa",
      "lora_target_modules": "model\\.language_model\\..*\\.(q_proj|k_proj|v_proj|o_proj|gate_proj|up_proj|down_proj)$",
      "memory": {
        "H200": {
          "max_padded_tokens": 8000,
          "provenance": "scratch/probe_batch_memory.py (commit 83343e7) on the 4xH200 pod; Slack #fellows-only-callum 2026-08-08"
        }
      }
    }
  },
  "dataset": {
    "repo": "dougalldeepmind/2026-09-15-da-gpt-fluff-removed-7-mix",
    "file": "mixture.jsonl",
    "revision": "5444733e6cd26222516cae093c33cd332822c7b2"
  },
  "reasoning_traces": {
    "model": "qwen/qwen3.6-27b",
    "family": "qwen36",
    "judge": "google/gemini-3-flash-preview",
    "sources": [
      "tulu3_if",
      "self_oss_instruct",
      "lima"
    ],
    "fraction": 0.5,
    "seed": 0,
    "rows": null,
    "turns": 1135,
    "provenance": "enrichment_report.json of dougalldeepmind/2026-09-08-nosynth-mix@7e991f58e86e (scratch/reasoning_backfill, pre-record)",
    "inherited_from": {
      "repo": "dougalldeepmind/2026-09-08-nosynth-mix",
      "revision": "7e991f58e86eff0b0a9f15a54ebeddfffb5b14dd"
    },
    "family_mismatch_allowed": false
  },
  "command": "scripts/train/train_lora.py --config configs/train/sft.yaml model=qwen36 data_repo=dougalldeepmind/2026-09-15-da-gpt-fluff-removed-7-mix data_file=mixture.jsonl data_revision=5444733e6cd26222516cae093c33cd332822c7b2 base_model_revision=6a9e13bd6fc8f0983b9b99948120bc37f49c13e9 seed=0 hf_repo=2026-09-15-qwen36-da-gpt-fluff-removed-0 wandb=true",
  "git_sha": "9a7d93b3e443b533e81098ba31d6b87149f5fce9",
  "timestamp": "20260915_190123"
}