# Shared References

Papers and documentation that apply to **more than one** of the three spam-classifier projects live in this folder. Project-specific references live inside each project's own `docs/references/` folder.

## How to browse

**Double-click [`how-to.html`](how-to.html)** to open the full reference index in your browser. It has clickable links to every local PDF and HTML guide, plus online fallbacks.

## What's in here

### Papers
- `papers/Attention_Vaswani2017.pdf` — The original Transformer paper
- `papers/LoRA_Hu2021.pdf` — LoRA fine-tuning technique (used by MLX + Liquid)
- `papers/QLoRA_Dettmers2023.pdf` — Quantized LoRA (applies to the MLX project's 4-bit model)
- `papers/PEFT_Survey.pdf` — Han 2024 survey of parameter-efficient fine-tuning methods

### Guides (HTML, open in browser)
- `guides/HuggingFace_LLM_FullFineTune_Tutorial.html` — LearnHuggingFace full fine-tune walkthrough
- `guides/HuggingFace_PEFT_Quicktour.html` — HF PEFT quickstart
- `guides/HuggingFace_PEFT_LoRA_Conceptual.html` — What LoRA `r`, `alpha`, `target_modules` mean
- `guides/HuggingFace_TRL_SFTTrainer.html` — TRL `SFTTrainer` reference (used by Liquid project)
- `guides/HuggingFace_ChatTemplates.html` — `tokenizer.apply_chat_template()` reference
- `guides/Unsloth_LoRA_Hyperparameters.html` — Practical LoRA hyperparameter advice

## Per-project references

Each project has its own folder with project-specific material (model tech reports, library docs, XAI papers, etc.):

| Project | Reference folder |
|---------|------------------|
| MLX | `../spam-classifier-mlx/docs/references/` |
| Liquid | `../spam-classifier-liquid/docs/references/` |
| XAI | `../spam-xai-project/docs/references/` |

Each contains its own `how-to.html`.

## Re-downloading or updating

If you want to refresh any of the downloaded files (e.g., HuggingFace updated their docs), re-run the fetch script from the LLM Project root:

```
bash fetch_references.sh
```

The script is safe to re-run — it overwrites existing files.