c4_1pct_9e18_seed1_2026-07-04_22-19-09_336791-pt

Trained with nanochat. Checkpoint at step 9,875.

W&B run: https://wandb.ai/alexksternteam/optimal_seed_replicas_v0/runs/vcq4f7c2

Outcome

metric	value
step	9875
smooth_train_loss	2.834476947784424
min_objective	0.8726849959751083
flops_used	8.91008145424384e+18
flops_per_token	3441950720.0
total_training_time	3083.8585999011993

Training config

{
  "model_pt": {
    "sequence_len": 2048,
    "vocab_size": 65536,
    "n_layer": 19,
    "n_head": 10,
    "n_kv_head": 10,
    "n_embd": 1280
  },
  "model_ppt": {
    "sequence_len": 2048,
    "vocab_size": 65536,
    "n_layer": 19,
    "n_head": 10,
    "n_kv_head": 10,
    "n_embd": 1280
  },
  "optim": {
    "matrix_lr": 0.015,
    "embedding_lr": 0.3,
    "unembedding_lr": 0.004,
    "weight_decay": 0.0
  },
  "train": {
    "num_iterations": 1000,
    "eval_every_n_steps": null,
    "eval_every_flops_frac": null,
    "save_every_n_steps": null,
    "log_every_n_steps": 100,
    "eval_at_end": true,
    "save_at_end": true,
    "grad_clip": 1.0,
    "ema_beta": 0.0
  },
  "eval": {
    "eval_steps": 16,
    "eval_tokens": 2016000
  },
  "hardware": {
    "device_batch_size": 16,
    "grad_accum_steps": 2,
    "peak_tflops": 2250.0
  },
  "wandb": {
    "enabled": true,
    "notes": "",
    "group": "9e18_ppt0.01",
    "tags": [
      "nanochat_gpt",
      "pt_fineweb-nanochatbpe-20B",
      "ppt_c4-nanochatbpe-10B",
      "seed_1",
      "case_b",
      "lr_trapezoid",
      "depth_19"
    ],
    "entity": null
  },
  "data_pt": "fineweb-nanochatbpe-20B",
  "data_ppt": "c4-nanochatbpe-10B",
  "extra_eval": [
    "c4-nanochatbpe-10B"
  ],
  "train_split": "train",
  "pad_vocab": 65536,
  "ppt_vocab_size": null,
  "ppt_same_vocab_as_pt": true,
  "ppt_unified_lr": true,
  "lr_kind": "trapezoid",
  "lr_warmup_ratio": 0.0,
  "lr_warmdown_ratio": 0.4,
  "lr_final_frac": 0.0,
  "ppt_lr_kind": "trapezoid",
  "ppt_lr_warmup_ratio": null,
  "ppt_lr_warmdown_ratio": null,
  "ppt_lr_final_frac": 0.0,
  "ppt_lr": null,
  "ppt_weight_decay": 0.0,
  "eval_tokens": 2016000,
  "reinit_embed_at_transition": false,
  "reset_optimizer_at_transition": false,
  "depth": 19,
  "compile_model": true,
  "model_project": "optimal_seed_replicas_v0",
  "run_name": "c4_1pct_9e18_seed1",
  "target_flops": 9e+18,
  "alpha_ppt": 0.01,
  "use_measured_flops": true,
  "flops_per_token": null,
  "ppt_flops_per_token": null,
  "seed": 1,
  "metrics_output_file": null,
  "push_to_hf": true,
  "hf_repo_org": "alexkstern",
  "cleanup_checkpoints": true
}

Files

• model_009875.pt — model weights (state_dict).
• meta_009875.json — training metadata.
• config_009875.json — run config snapshot.
• rng_009875.pt — RNG state (when present).