{
  "created_at": "2026-07-16 23:52:50",
  "name": "v52_public091_full_lr2e5_ep1",
  "base": "OpenOneRec/OneReason-0.8B-pretrain-competition",
  "dataset": "v49_public_091_exact",
  "dataset_manifest": {
    "path": "/home/ll/llm4rec/experiments/data/v49_public_091_exact.jsonl",
    "records": 32705,
    "groups": {
      "rec": 18651,
      "item": 9684,
      "user": 2792,
      "world": 1578
    },
    "routes": {
      "no_think": 20934,
      "think": 11771
    },
    "thought_modes": {
      "empty": 20934,
      "filled": 11771
    },
    "sha256": "cdedea13c560d3453697f4a3c9b96a2303bd708c84833034fa953d16482b9925",
    "source_path": "/home/ll/llm4rec/data/external/kuaishou-llmrec-sft-baseline-0.91/train.jsonl",
    "source_url": "https://huggingface.co/datasets/Frinkleko/kuaishou-llmrec-sft-baseline-0.91/resolve/main/train.jsonl?download=true",
    "source_sha256": "4d6b29d76974c9a1517c1b583858e744cae019cb26e1e2d90066e237ebbcf5f8",
    "source_license": "Apache-2.0",
    "transform": "Lossless field mapping from system/prompt/response to instruction/input/output; original seed-42 order retained."
  },
  "learning_rate": "2.0e-5",
  "epochs": 1.0,
  "method": "full",
  "weight_decay": 0.0,
  "validation_size": 0.0,
  "lora": {
    "rank": null,
    "alpha": null,
    "dropout": null,
    "target": null
  },
  "stages": null,
  "note": "Full-SFT counterpart using v7's successful 2e-5 scale, now on balanced CoT/direct clean data instead of final-only data.",
  "evaluation": null,
  "cot_policy": "Preserve meaningful CoT for /think and pure final outputs for /no_think.",
  "checkpoint_policy": "save_strategy=no; only the final model is retained.",
  "script": "/home/ll/llm4rec/experiments/run_experiments.py",
  "script_sha256": "bc57a4d1337c7182996ece63ddc844185043b68ae2a2dbe793e3969bbccc2e2d",
  "data_builder": "/home/ll/llm4rec/experiments/source_data.py",
  "data_builder_sha256": "a1216b14c8c516f9fde2fcaab9d70ec5b8a200033fbd1ee330c6b461213587ad",
  "data_cleaner": null,
  "data_cleaner_sha256": null,
  "configs": [
    "/home/ll/llm4rec/experiments/configs/v52_public091_full_lr2e5_ep1.yaml"
  ],
  "reproduce": "/home/ll/llm4rec/demo/LLaMA-Factory/.venv/bin/python /home/ll/llm4rec/experiments/run_experiments.py --single v52_public091_full_lr2e5_ep1 --gpu <gpu>",
  "training_result": {
    "name": "v52_public091_full_lr2e5_ep1",
    "gpu": 1,
    "started_at": "2026-07-17 01:41:03",
    "status": "ok",
    "wall_seconds": 2611.22,
    "finished_at": "2026-07-17 02:24:34",
    "epoch": 1.0,
    "total_flos": 1.563295846805975e+17,
    "train_loss": 1.6355595296145948,
    "train_runtime": 2594.717,
    "train_samples_per_second": 0.503,
    "train_steps_per_second": 0.126,
    "global_step": 326,
    "first_logged_loss": 2.659077453613281,
    "last_logged_loss": 1.512490463256836,
    "min_logged_loss": 1.5008227348327636,
    "tail_loss_mean": 1.5448276201883953,
    "logged_loss_change": -1.1465869903564452,
    "batch": "v49-v53"
  }
}