README.md

---
language: en
license: mit
tags:
  - sdf
  - classification
  - qwen2.5
  - gguf
  - content-type
  - web-content
base_model: Qwen/Qwen2.5-1.5B-Instruct
pipeline_tag: text-generation
---

# SDF Classify

Content type classifier for the [SDF Protocol](https://sdfprotocol.org). Fine-tuned from Qwen2.5-1.5B-Instruct using QLoRA.

## Purpose

Classifies web content into SDF's hierarchical type system: 10 parent types and 50+ subtypes (e.g., `article.news`, `commerce.product`, `documentation.api_docs`).

## Training

- **Base model**: Qwen2.5-1.5B-Instruct
- **Method**: QLoRA (rank 32, alpha 64, dropout 0.05)
- **Training data**: 2,335 classified web documents
- **Accuracy**: 95.2% exact type match

## Files

| File | Size | Description |
|------|------|-------------|
| `sdf-classify-Qwen2.5-1.5B-Instruct-Q4_K_M.gguf` | 941 MB | Quantized (Q4_K_M) — recommended for deployment |
| `sdf-classify-Qwen2.5-1.5B-Instruct-f16.gguf` | 2.9 GB | Full precision (f16) |
| `Modelfile` | — | Ollama import configuration |

## Usage with Ollama

```bash
# Download the Q4_K_M file, then:
ollama create sdf-classify -f Modelfile
```

## Part of SDF Protocol

- **Protocol**: [sdfprotocol.org](https://sdfprotocol.org)
- **Specification**: [github.com/sdfprotocol/sdf](https://github.com/sdfprotocol/sdf)
- **Whitepaper**: [DOI 10.5281/zenodo.18559223](https://doi.org/10.5281/zenodo.18559223)
- **Extractor model**: [pranab2050/sdf-extract](https://huggingface.co/pranab2050/sdf-extract)

## Citation

```bibtex
@article{sarkar2026sdf,
  title={Convert Once, Consume Many: SDF for Cacheable, Typed Semantic Extraction from Web Pages},
  author={Sarkar, Pranab},
  year={2026},
  doi={10.5281/zenodo.18559223},
  publisher={Zenodo}
}
```