---
name: hf-docs
description: "Search and fetch current Hugging Face documentation for Transformers, TRL, PEFT, Datasets, Trackio, and other libraries."
disable-model-invocation: false
---

# hf-docs — Hugging Face Documentation

## Purpose

Look up current API usage patterns, trainer configs, and library documentation before writing ML code. Do not rely on memory — HF library APIs change frequently.

## Tools

- `hf_doc_search`: Search HF documentation by product and query.
- `hf_doc_fetch`: Fetch the full content of a specific doc page by URL.

## Products Available

| Product | Use For |
|---|---|
| transformers | Model loading, tokenizers, trainers, inference |
| trl | SFT, DPO, GRPO, PPO trainers |
| peft | LoRA, adapters, parameter-efficient fine-tuning |
| datasets | Dataset loading, streaming, processing |
| accelerate | Distributed training, mixed precision |
| hub | Repo operations, auth, upload/download |
| trackio | Experiment tracking, metrics, dashboards |
| sentence_transformers | Embedding models, training, similarity |
| diffusers | Diffusion pipelines, fine-tuning |
| optimum | Hardware optimization, model export |

## Workflow

1. Use `hf_doc_search` to find the most relevant current page for the exact library or trainer.
2. Use `hf_doc_fetch` to read the full content of the page before coding.
3. Extract the exact imports, class names, config parameters, and argument names.
4. Cross-check any example against the current library version you are targeting.
5. Use the current API in your implementation, not memory or old snippets.

## Example

```
_hf_doc_search(product="trl", query="SFTConfig parameters")
_hf_doc_fetch(url="https://huggingface.co/docs/trl/sft_trainer")
```

## Docs-First Rule

Before writing any training, fine-tuning, inference, or evaluation code:
- Find at least one current working implementation pattern from HF docs or a relevant repo.
- Verify import paths, trainer class names, and config field names.
- Check that the example matches your library version constraints.
- If the API changed recently, prefer the docs page over older code examples.