Hugging Face's logo

alexkstern
/
code_5pct_1e18_seed1_2026-07-12_01-12-46_897145-pt

nanochat
nanochat_gpt
pt_fineweb-nanochatbpe-20B
ppt_github-code-nanochatbpe-1B
seed_1
case_b
lr_trapezoid
depth_12
Model card Files
xet
 Community
code_5pct_1e18_seed1_2026-07-12_01-12-46_897145-pt
File size: 3,174 Bytes
096a6fb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
---
library_name: nanochat
license: apache-2.0
tags:
- nanochat
- nanochat_gpt
- pt_fineweb-nanochatbpe-20B
- ppt_github-code-nanochatbpe-1B
- seed_1
- case_b
- lr_trapezoid
- depth_12
---

# code_5pct_1e18_seed1_2026-07-12_01-12-46_897145-pt

Trained with [nanochat](https://github.com/karpathy/nanochat). Checkpoint at step **3,368**.

**W&B run:** https://wandb.ai/alexksternteam/optimal_seed_replicas_v1/runs/33ea3a6t

## Outcome

| metric | value |
| --- | --- |
| `step` | 3368 |
| `smooth_train_loss` | 3.2399237155914307 |
| `min_objective` | 0.9851442932664687 |
| `flops_used` | 9.498592991426642e+17 |
| `flops_per_token` | 1075838976.0 |
| `total_training_time` | 422.86022329330444 |

## Training config

```json
{
  "model_pt": {
    "sequence_len": 2048,
    "vocab_size": 65536,
    "n_layer": 12,
    "n_head": 6,
    "n_kv_head": 6,
    "n_embd": 768
  },
  "model_ppt": {
    "sequence_len": 2048,
    "vocab_size": 65536,
    "n_layer": 12,
    "n_head": 6,
    "n_kv_head": 6,
    "n_embd": 768
  },
  "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": 8,
    "eval_tokens": 2016000
  },
  "hardware": {
    "device_batch_size": 32,
    "grad_accum_steps": 1,
    "peak_tflops": 2250.0
  },
  "wandb": {
    "enabled": true,
    "notes": "",
    "group": "1e18_ppt0.05",
    "tags": [
      "nanochat_gpt",
      "pt_fineweb-nanochatbpe-20B",
      "ppt_github-code-nanochatbpe-1B",
      "seed_1",
      "case_b",
      "lr_trapezoid",
      "depth_12"
    ],
    "entity": null
  },
  "data_pt": "fineweb-nanochatbpe-20B",
  "data_ppt": "github-code-nanochatbpe-1B",
  "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,
  "ppt_device_batch_size": null,
  "ppt_grad_accum_steps": null,
  "eval_tokens": 2016000,
  "reinit_embed_at_transition": false,
  "reset_optimizer_at_transition": false,
  "depth": 12,
  "compile_model": true,
  "model_project": "optimal_seed_replicas_v1",
  "run_name": "code_5pct_1e18_seed1",
  "target_flops": 1e+18,
  "alpha_ppt": 0.05,
  "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_003368.pt` — model weights (`state_dict`).
- `meta_003368.json` — training metadata.
- `config_003368.json` — run config snapshot.
- `rng_003368.pt` — RNG state (when present).