2b/2b_parent_sgd/config.json

{
 "run_id": "2b_parent_sgd",
 "model_id": "Qwen/Qwen3-0.6B-Base",
 "task": "gsm8k",
 "seq_len": 512,
 "batch_size": 8,
 "grad_accum": 1,
 "max_steps": 600,
 "lora_r": 8,
 "lora_alpha": null,
 "targets": [
  "q_proj",
  "k_proj",
  "v_proj",
  "o_proj"
 ],
 "optimizer": "sgd",
 "lr": 0.001,
 "weight_decay": 0.01,
 "betas": [
  0.9,
  0.999
 ],
 "grad_clip": 1.0,
 "snapshot_every": 10,
 "probe_every": 10,
 "eval_every": 100,
 "eval_batches": 8,
 "seed": 17,
 "data_seed": 17,
 "warm_start": null,
 "gauge_transform": "none",
 "gauge_c": 4.0,
 "gauge_seed": 0,
 "save_warm_at": 100,
 "device": "auto",
 "base_dtype": "auto",
 "out_root": "runs"
}