scripts/TRAIN.md

# Train

## Tokenizer

```bash
cd scripts
python -m venv venv
source venv/bin/activate
pip install -U -r requirements.in
```

```bash
python -B train_tokenizer.py
```

## Dataset

```bash
cd scripts
python -m venv venv-lit
source venv-lit/bin/activate
pip install -U -r requirements-lit.in
```

```bash
python -B prepare_pretrain_dataset.py
```

## Model

```bash
cd scripts
python -m venv venv-lit
source venv-lit/bin/activate
pip install -U -r requirements-lit.in
```

```bash
litgpt pretrain --config ./model.yaml
```

```bash
litgpt convert_from_litgpt out/pretrain/final/ out/converted_model
cp config.json out/pretrain/final/
cp config.json out/converted_model/
```

```python
import torch
from transformers import AutoModel

state_dict = torch.load('out/converted_model/model.pth')
model = AutoModel.from_pretrained('TinyLlama/TinyLlama_v1.1', state_dict=state_dict, ignore_mismatched_sizes=True)
model.save_pretrained('out/converted_model/')
```

## Evaluate

```bash
# litgpt evaluate --tasks 'hellaswag,gsm8k,truthfulqa_mc2,mmlu,winogrande,arc_challenge' --batch_size 8 out/pretrain/final/

litgpt evaluate --tasks 'hellaswag,gsm8k,truthfulqa_mc2,mmlu,mmlu_pro,winogrande,arc_challenge,leaderboard,ifeval,mgsm_direct,mathqa,gpqa' --batch_size 8 out/pretrain/final/
```