# Supervised Learning

## SFT on NoRobots

```bash
python -m tinker_cookbook.recipes.chat_sl.train \
    model_name=Qwen/Qwen3.5-9B-Base \
    dataset=no_robots \
    learning_rate=5e-4 \
    batch_size=64 \
    lora_rank=64 \
    eval_every=20 \
    save_every=20 \
    wandb_project=cookbook_sl
```

After 140 steps of training, `test/nll` decreases to 1.663.

## SFT on Tulu3 dataset

```bash
python -m tinker_cookbook.recipes.chat_sl.train \
    model_name=Qwen/Qwen3.5-9B-Base \
    dataset=tulu3 \
    learning_rate=5e-4 \
    batch_size=512 \
    lora_rank=64 \
    eval_every=500 \
    save_every=500 \
    wandb_project=cookbook_sl
```

After 1800 steps of training, `test/nll` decreases to ~0.645. Performance can be further improved by training longer with a higher `lora_rank` and lower `batch_size`.

## Adding your own dataset

The base classes in [tinker_cookbook/supervised/data.py](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/supervised/data.py) support loading new data in the following way:

- `SupervisedDatasetFromHFDataset` loads dataset on Hugging Face hub with a postprocessing function
- `StreamingSupervisedDatasetFromHFDataset` works similarly, but supports streaming
- `FromConversationFileBuilder` supports data loading from a JSONL file

## Sweep results

See [results/sft_sweep.md](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/recipes/chat_sl/results/sft_sweep.md) for empirical hyperparameter sweep results (learning rate, LoRA rank) across multiple models on the tulu3 dataset.