training_metadata.json

{
  "name": "gcb_doc_clean",
  "seed": 0,
  "attn": "flex",
  "compile": true,
  "seq_len": 8192,
  "micro_batch": 4,
  "grad_accum": 2,
  "lr": 0.0003,
  "wd": 0.1,
  "clip": 1.0,
  "warmup_frac": 0.04,
  "decay_frac": 0.4,
  "tier_weights": [
    1.0,
    1.0,
    0.3
  ],
  "lam": 0.1,
  "total_steps": 120000,
  "hard_max_steps": 1000000,
  "num_workers": 16,
  "log_every": 20,
  "ckpt_every": 2000,
  "eval_every_ckpt": true,
  "eval_n": 256,
  "out_dir": "gcb_doc_clean",
  "anneal_phases": [
    [
      0.5,
      [
        1.0,
        1.0,
        0.3
      ]
    ],
    [
      0.9,
      [
        1.0,
        1.0,
        0.0
      ]
    ],
    [
      1.0,
      [
        1.0,
        0.0,
        0.0
      ]
    ]
  ],
  "_source_checkpoint": "best.pt",
  "_source_step": 116001
}