Views
No views yet
| metric | value |
|---|---|
step | 17950 |
smooth_train_loss | 2.544304609298706 |
min_objective | 0.8280173380890073 |
flops_used | 2.6999167727016346e+19 |
flops_per_token | 5737807872.0 |
total_training_time | 8242.587728977203 |
1{
2 "model_pt": {
3 "sequence_len": 2048,
4 "vocab_size": 65536,
5 "n_layer": 24,
6 "n_head": 12,
7 "n_kv_head": 12,
8 "n_embd": 1536
9 },
10 "model_ppt": null,
11 "optim": {
12 "matrix_lr": 0.012,
13 "embedding_lr": 0.3,
14 "unembedding_lr": 0.004,
15 "weight_decay": 0.0
16 },
17 "train": {
18 "num_iterations": 1000,
19 "eval_every_n_steps": null,
20 "eval_every_flops_frac": null,
21 "save_every_n_steps": null,
22 "save_at_steps": null,
23 "log_every_n_steps": 100,
24 "eval_at_end": true,
25 "save_at_end": true,
26 "grad_clip": 1.0,
27 "ema_beta": 0.0
28 },
29 "eval": {
30 "eval_steps": 8,
31 "eval_tokens": 2016000
32 },
33 "hardware": {
34 "device_batch_size": 32,
35 "grad_accum_steps": 1,
36 "peak_tflops": 2250.0
37 },
38 "wandb": {
39 "enabled": true,
40 "notes": "",
41 "group": "3e19_vanilla",
42 "tags": [
43 "nanochat_gpt",
44 "pt_fineweb-nanochatbpe-20B",
45 "no_ppt",
46 "seed_1",
47 "single",
48 "lr_trapezoid",
49 "depth_24"
50 ],
51 "entity": null
52 },
53 "data_pt": "fineweb-nanochatbpe-20B",
54 "data_ppt": null,
55 "extra_eval": [
56 "c4-nanochatbpe-10B"
57 ],
58 "train_split": "train",
59 "pad_vocab": 65536,
60 "ppt_vocab_size": null,
61 "ppt_same_vocab_as_pt": false,
62 "ppt_unified_lr": false,
63 "lr_kind": "trapezoid",
64 "lr_warmup_ratio": 0.0,
65 "lr_warmdown_ratio": 0.5,
66 "lr_final_frac": 0.0,
67 "ppt_lr_kind": "trapezoid",
68 "ppt_lr_warmup_ratio": null,
69 "ppt_lr_warmdown_ratio": null,
70 "ppt_lr_final_frac": 0.0,
71 "ppt_lr": null,
72 "ppt_weight_decay": 0.0,
73 "ppt_device_batch_size": null,
74 "ppt_grad_accum_steps": null,
75 "eval_tokens": 2016000,
76 "reinit_embed_at_transition": true,
77 "moment_match_embed_reinit": false,
78 "reset_optimizer_at_transition": true,
79 "random_pt_head": false,
80 "depth": 24,
81 "compile_model": true,
82 "model_project": "vanilla_pred_27e18_seeds_v0",
83 "run_name": "vanilla_pred_27e18_d24_seed1",
84 "target_flops": 2.7e+19,
85 "alpha_ppt": 0.0,
86 "use_measured_flops": true,
87 "flops_per_token": null,
88 "ppt_flops_per_token": null,
89 "seed": 1,
90 "metrics_output_file": null,
91 "push_to_hf": true,
92 "hf_repo_org": "alexkstern",
93 "cleanup_checkpoints": true
94}model_017950.pt — model weights (state_dict).meta_017950.json — training metadata.config_017950.json — run config snapshot.rng_017950.pt — RNG state (when present).