This commit is contained in:
ob1 2025-10-28 14:25:24 +08:00
parent ca17179ec2
commit 111f7cf9de

309
run4_w_rl.md Normal file
View File

@ -0,0 +1,309 @@
# nanochat training report
Generated: 2025-10-24 13:49:50
## Environment
### Git Information
- Branch: master
- Commit: ec11d39 (dirty)
- Message: rename
### Hardware
- Platform: Linux
- CPUs: 112 cores (224 logical)
- Memory: 2015.6 GB
- GPUs: 8x NVIDIA H100 80GB HBM3
- GPU Memory: 632.8 GB total
- CUDA Version: 12.8
- Hourly Rate: $24.00/hour
### Software
- Python: 3.10.12
- PyTorch: 2.8.0+cu128
### Bloat
- Characters: 353,214
- Lines: 8,982
- Files: 45
- Tokens (approx): 88,303
- Dependencies (uv.lock lines): 2,004
Run started: 2025-10-24 13:50:00
---
## Tokenizer training
timestamp: 2025-10-24 13:52:13
- max_chars: 2,000,000,000
- doc_cap: 10,000
- vocab_size: 65,536
- train_time: 77.3423
- num_special_tokens: 9
- token_bytes_min: 1
- token_bytes_max: 32
- token_bytes_mean: 6.9151
- token_bytes_std: 2.8736
## Tokenizer evaluation
timestamp: 2025-10-24 13:52:25
### Comparison with GPT-2
| Text Type | Bytes | GPT-2 Tokens | GPT-2 Ratio | Ours Tokens | Ours Ratio | Relative Diff % |
|-----------|-------|--------------|--------------|-------------|------------|-----------------|
| news | 1819 | 404 | 4.50 | 375 | 4.85 | +7.2% |
| korean | 893 | 745 | 1.20 | 721 | 1.24 | +3.2% |
| code | 1259 | 576 | 2.19 | 493 | 2.55 | +14.4% |
| math | 1834 | 936 | 1.96 | 966 | 1.90 | -3.2% |
| science | 1112 | 260 | 4.28 | 225 | 4.94 | +13.5% |
| fwe-train | 4208518 | 900364 | 4.67 | 856901 | 4.91 | +4.8% |
| fwe-val | 4908443 | 1059062 | 4.63 | 1010356 | 4.86 | +4.6% |
### Comparison with GPT-4
| Text Type | Bytes | GPT-4 Tokens | GPT-4 Ratio | Ours Tokens | Ours Ratio | Relative Diff % |
|-----------|-------|--------------|--------------|-------------|------------|-----------------|
| news | 1819 | 387 | 4.70 | 375 | 4.85 | +3.1% |
| korean | 893 | 364 | 2.45 | 721 | 1.24 | -98.1% |
| code | 1259 | 309 | 4.07 | 493 | 2.55 | -59.5% |
| math | 1834 | 832 | 2.20 | 966 | 1.90 | -16.1% |
| science | 1112 | 249 | 4.47 | 225 | 4.94 | +9.6% |
| fwe-train | 4208518 | 874799 | 4.81 | 856901 | 4.91 | +2.0% |
| fwe-val | 4908443 | 1029691 | 4.77 | 1010356 | 4.86 | +1.9% |
## Base model training
timestamp: 2025-10-24 17:18:28
- run: run4_w_rl
- depth: 20
- max_seq_len: 2048
- num_iterations: -1
- target_flops: -1.0000
- target_param_data_ratio: 20
- device_batch_size: 32
- total_batch_size: 524,288
- embedding_lr: 0.2000
- unembedding_lr: 0.0040
- weight_decay: 0.0000
- matrix_lr: 0.0200
- grad_clip: 1.0000
- eval_every: 250
- eval_tokens: 10,485,760
- core_metric_every: 2000
- core_metric_max_per_task: 500
- sample_every: 2000
- model_tag:
- Number of parameters: 560,988,160
- Number of FLOPs per token: 3.491758e+09
- Calculated number of iterations: 21,400
- Number of training tokens: 11,219,763,200
- Tokens : Params ratio: 20.0000
- DDP world size: 8
- warmup_ratio: 0.0000
- warmdown_ratio: 0.2000
- final_lr_frac: 0.0000
- Minimum validation bpb: 0.8118
- Final validation bpb: 0.8118
- CORE metric estimate: 0.2102
- MFU %: 47.96%
- Total training flops: 3.917670e+19
- Total training time: 189.59m
- Peak memory usage: 75422.02MiB
## Base model loss
timestamp: 2025-10-24 17:22:16
- train bpb: 0.8147
- val bpb: 0.8119
- sample 0: <|bos|>The capital of France is Paris. The capital of the United States is Washington, D.C. The capital
- sample 1: <|bos|>The chemical symbol of gold is Au. It is a soft, malleable, ductile, and ductile metal. It
- sample 2: <|bos|>If yesterday was Friday, then tomorrow will be Monday. If yesterday was Monday, then tomorrow will be Tuesday. If yesterday was
- sample 3: <|bos|>The opposite of hot is cold, and the opposite of cold is hot. The opposite of hot is cold
- sample 4: <|bos|>The planets of the solar system are: Mercury, Venus, Earth, Mars, Jupiter, Saturn, Uranus, Neptune,
- sample 5: <|bos|>My favorite color is red. I love red. I love red. I love red. I love
- sample 6: <|bos|>If 5*x + 3 = 13, then x is 3. If 5*x + 3 = 13, then
## Base model evaluation
timestamp: 2025-10-24 17:26:29
- Model: base_model (step 21400)
- CORE metric: 0.2020
- hellaswag_zeroshot: 0.2588
- jeopardy: 0.0666
- bigbench_qa_wikidata: 0.5234
- arc_easy: 0.5292
- arc_challenge: 0.1035
- copa: 0.2600
- commonsense_qa: 0.1605
- piqa: 0.3602
- openbook_qa: 0.0960
- lambada_openai: 0.3757
- hellaswag: 0.2606
- winograd: 0.2747
- winogrande: 0.0766
- bigbench_dyck_languages: 0.1350
- agi_eval_lsat_ar: 0.0272
- bigbench_cs_algorithms: 0.3833
- bigbench_operators: 0.1571
- bigbench_repeat_copy_logic: 0.0000
- squad: 0.2170
- coqa: 0.2053
- boolq: -0.2128
- bigbench_language_identification: 0.1868
## Midtraining
timestamp: 2025-10-24 17:44:33
- run: run4_w_rl
- dtype: bfloat16
- max_seq_len: 2048
- device_batch_size: 32
- unembedding_lr: 0.0040
- embedding_lr: 0.2000
- matrix_lr: 0.0200
- init_lr_frac: 1.0000
- weight_decay: 0.0000
- final_lr_frac: 0.0000
- eval_every: 150
- eval_tokens: 10,485,760
- total_batch_size: 524,288
- Number of iterations: 765
- DDP world size: 8
- Minimum validation bpb: 0.4152
## Chat evaluation mid
timestamp: 2025-10-24 17:50:53
- source: mid
- task_name: None
- dtype: bfloat16
- temperature: 0.0000
- max_new_tokens: 512
- num_samples: 1
- top_k: 50
- batch_size: 8
- model_tag: None
- step: None
- max_problems: None
- ARC-Easy: 0.3851
- ARC-Challenge: 0.3089
- MMLU: 0.3123
- GSM8K: 0.0281
- HumanEval: 0.0732
- ChatCORE metric: 0.0886
## Chat SFT
timestamp: 2025-10-24 17:57:11
- run: run4_w_rl
- source: mid
- dtype: bfloat16
- device_batch_size: 4
- num_epochs: 1
- max_iterations: -1
- target_examples_per_step: 32
- unembedding_lr: 0.0040
- embedding_lr: 0.2000
- matrix_lr: 0.0200
- weight_decay: 0.0000
- init_lr_frac: 0.0200
- eval_every: 100
- eval_steps: 100
- eval_metrics_every: 200
- Training rows: 20,843
- Number of iterations: 651
- Training loss: 1.1895
- Validation loss: 1.0660
## Chat evaluation sft
timestamp: 2025-10-24 18:02:19
- source: sft
- task_name: None
- dtype: bfloat16
- temperature: 0.0000
- max_new_tokens: 512
- num_samples: 1
- top_k: 50
- batch_size: 8
- model_tag: None
- step: None
- max_problems: None
- ARC-Easy: 0.4230
- ARC-Challenge: 0.3114
- MMLU: 0.3232
- GSM8K: 0.0553
- HumanEval: 0.0671
- ChatCORE metric: 0.1065
## Chat RL
timestamp: 2025-10-24 18:49:25
- run: run4_w_rl
- source: sft
- dtype: bfloat16
- device_batch_size: 8
- examples_per_step: 16
- num_samples: 16
- max_new_tokens: 256
- temperature: 1.0000
- top_k: 50
- unembedding_lr: 0.0040
- embedding_lr: 0.2000
- matrix_lr: 0.0200
- weight_decay: 0.0000
- init_lr_frac: 0.0500
- num_epochs: 1
- save_every: 60
- eval_every: 60
- eval_examples: 400
## Chat evaluation rl
timestamp: 2025-10-24 18:51:57
- source: rl
- task_name: GSM8K
- dtype: bfloat16
- temperature: 0.0000
- max_new_tokens: 512
- num_samples: 1
- top_k: 50
- batch_size: 8
- model_tag: None
- step: None
- max_problems: None
- GSM8K: 0.0751
## Summary
- Characters: 353,214
- Lines: 8,982
- Files: 45
- Tokens (approx): 88,303
- Dependencies (uv.lock lines): 2,004
| Metric | BASE | MID | SFT | RL |
|-----------------|----------|----------|----------|----------|
| CORE | 0.2020 | - | - | - |
| ARC-Challenge | - | 0.3089 | 0.3114 | - |
| ARC-Easy | - | 0.3851 | 0.4230 | - |
| GSM8K | - | 0.0281 | 0.0553 | 0.0751 |
| HumanEval | - | 0.0732 | 0.0671 | - |
| MMLU | - | 0.3123 | 0.3232 | - |
| ChatCORE | - | 0.0886 | 0.1065 | - |
Total wall clock time: 4h12m