{
 "run": "ocrvlm_v12_clean_q35",
 "model": "Qwen/Qwen3.5-2B",
 "steps": 2000,
 "max_steps": 2000,
 "samples": 32000,
 "best_checkpoint": "/opt/sansar/runs/ocrvlm_v12_clean_q35/checkpoint-2000",
 "best_val_loss": 0.06209864094853401,
 "evals": [
  [
   200,
   0.15769308805465698
  ],
  [
   400,
   0.1268068104982376
  ],
  [
   600,
   0.10351545363664627
  ],
  [
   800,
   0.09406991302967072
  ],
  [
   1000,
   0.08291669189929962
  ],
  [
   1200,
   0.07203876972198486
  ],
  [
   1400,
   0.06750340759754181
  ],
  [
   1600,
   0.06383494287729263
  ],
  [
   1800,
   0.062389373779296875
  ],
  [
   2000,
   0.06209864094853401
  ]
 ],
 "train_loss": 0.053184996942679086,
 "wall_seconds": 7254.4,
 "tokens_seen": 80725777,
 "peak_mem_gb": 34.63,
 "gpu": "NVIDIA A100-SXM4-40GB",
 "args": {
  "model": "Qwen/Qwen3.5-2B",
  "data_root": "/opt/sansar/ocrdata",
  "out_dir": "/opt/sansar/runs/ocrvlm_v12_clean_q35",
  "max_pixels": 1800000,
  "max_len": 6144,
  "samples": 30000,
  "max_steps": 2000,
  "grad_accum": 16,
  "lr": 1.5e-05,
  "vision_lr": 5e-06,
  "warmup_ratio": 0.03,
  "final_gold_frac": 0.15,
  "eval_steps": 200,
  "val_max": 300,
  "val_per_source": 120,
  "save_total_limit": 2,
  "workers": 8,
  "seed": 0,
  "freeze_vision": false,
  "optim": "adamw8bit",
  "no_grad_ckpt": false,
  "smoke": 0,
  "device_map": null,
  "pilot": false,
  "weights": "gold_twin=1.5,textbooks=0.5,air=0.2,synthetic=0.3"
 },
 "batch_size": 16
}