Views
No views yet
| metric | value |
|---|---|
step | 275 |
smooth_train_loss | 2.775472640991211 |
min_objective | 1.7841455652069105 |
flops_used | 1.499733860286464e+17 |
flops_per_token | 2080374784.0 |
total_training_time | 51.19044303894043 |
1{
2 "model_pt": {
3 "sequence_len": 2048,
4 "vocab_size": 65536,
5 "n_layer": 16,
6 "n_head": 8,
7 "n_kv_head": 8,
8 "n_embd": 1024
9 },
10 "model_ppt": {
11 "sequence_len": 2048,
12 "vocab_size": 65536,
13 "n_layer": 16,
14 "n_head": 8,
15 "n_kv_head": 8,
16 "n_embd": 1024
17 },
18 "optim": {
19 "matrix_lr": 0.015,
20 "embedding_lr": 0.3,
21 "unembedding_lr": 0.004,
22 "weight_decay": 0.0
23 },
24 "train": {
25 "num_iterations": 1000,
26 "eval_every_n_steps": null,
27 "eval_every_flops_frac": null,
28 "save_every_n_steps": null,
29 "log_every_n_steps": 100,
30 "eval_at_end": true,
31 "save_at_end": true,
32 "grad_clip": 1.0,
33 "ema_beta": 0.0
34 },
35 "eval": {
36 "eval_steps": 8,
37 "eval_tokens": 2016000
38 },
39 "hardware": {
40 "device_batch_size": 32,
41 "grad_accum_steps": 1,
42 "peak_tflops": 2250.0
43 },
44 "wandb": {
45 "enabled": true,
46 "notes": "",
47 "group": "3e18_ppt0.05",
48 "tags": [
49 "nanochat_gpt",
50 "pt_fineweb-nanochatbpe-20B",
51 "ppt_github-code-nanochatbpe-1B",
52 "seed_2",
53 "case_b",
54 "lr_trapezoid",
55 "depth_16"
56 ],
57 "entity": null
58 },
59 "data_pt": "fineweb-nanochatbpe-20B",
60 "data_ppt": "github-code-nanochatbpe-1B",
61 "extra_eval": [
62 "c4-nanochatbpe-10B"
63 ],
64 "train_split": "train",
65 "pad_vocab": 65536,
66 "ppt_vocab_size": null,
67 "ppt_same_vocab_as_pt": true,
68 "ppt_unified_lr": true,
69 "lr_kind": "trapezoid",
70 "lr_warmup_ratio": 0.02,
71 "lr_warmdown_ratio": 0.5,
72 "lr_final_frac": 0.0,
73 "ppt_lr_kind": "trapezoid",
74 "ppt_lr_warmup_ratio": null,
75 "ppt_lr_warmdown_ratio": null,
76 "ppt_lr_final_frac": 0.0,
77 "ppt_lr": null,
78 "ppt_weight_decay": 0.0,
79 "ppt_device_batch_size": null,
80 "ppt_grad_accum_steps": null,
81 "eval_tokens": 2016000,
82 "reinit_embed_at_transition": false,
83 "reset_optimizer_at_transition": false,
84 "depth": 16,
85 "compile_model": true,
86 "model_project": "optimal_seed_replicas_v1",
87 "run_name": "code_5pct_3e18_seed2",
88 "target_flops": 3e+18,
89 "alpha_ppt": 0.05,
90 "use_measured_flops": true,
91 "flops_per_token": null,
92 "ppt_flops_per_token": null,
93 "seed": 2,
94 "metrics_output_file": null,
95 "push_to_hf": true,
96 "hf_repo_org": "alexkstern",
97 "cleanup_checkpoints": true
98}model_000275.pt — model weights (state_dict).meta_000275.json — training metadata.config_000275.json — run config snapshot.rng_000275.pt — RNG state (when present).