train_stats.json

{
 "base": "Qwen/Qwen3-32B",
 "method": "fsdp2-custom-loop",
 "n_rows": 40560,
 "n_self": 0,
 "n_sequences": 20265,
 "n_tokens": 32377321,
 "world": 8,
 "micro_batch": 4,
 "accum": 2,
 "steps": 316,
 "lr": 1e-05,
 "optim": "AdamW fused, fp32 master, bf16 compute, fp32 grad reduce",
 "loss_curve": [
  2.1481621265411377,
  1.9153783321380615,
  1.699304223060608,
  1.5891406536102295,
  1.4919301271438599,
  1.4694232940673828,
  1.5090442895889282,
  1.463268756866455,
  1.473642349243164,
  1.3800721168518066,
  1.4203133583068848,
  1.3800207376480103,
  1.4587819576263428,
  1.4176052808761597,
  1.437802791595459,
  1.4048199653625488,
  1.4004789590835571,
  1.3153928518295288,
  1.3857331275939941,
  1.425294280052185,
  1.3739887475967407,
  1.42657470703125,
  1.3883891105651855,
  1.3731762170791626,
  1.4036970138549805,
  1.450828194618225,
  1.42012357711792,
  1.3541591167449951,
  1.4580155611038208,
  1.4323490858078003,
  1.3500598669052124,
  1.3436286449432373,
  1.2824114561080933,
  1.4008318185806274,
  1.292543649673462,
  1.3753026723861694,
  1.3455853462219238,
  1.3969128131866455,
  1.4189624786376953,
  1.4118071794509888,
  1.2936642169952393,
  1.340247631072998,
  1.344330072402954,
  1.3777103424072266,
  1.404588222503662,
  1.3995976448059082,
  1.391228437423706,
  1.3635567426681519,
  1.3644399642944336,
  1.3421406745910645,
  1.2756061553955078,
  1.3820265531539917,
  1.277800440788269,
  1.3787689208984375,
  1.3514537811279297,
  1.3293616771697998,
  1.4152438640594482,
  1.3695340156555176,
  1.3992278575897217,
  1.3072197437286377,
  1.3011447191238403,
  1.3627418279647827,
  1.2566815614700317,
  1.3739030361175537
 ],
 "loss_first": [
  2.1481621265411377
 ],
 "loss_last": [
  1.3739030361175537
 ],
 "train_seconds": 4039.421051263809,
 "tokens_per_second": 8015.336997332859,
 "resumed_from": null
}