---
base_model: TinyLlama/TinyLlama-1.1B-Chat-v1.0
datasets:
  - ducklin404/chezz_dataset
library_name: peft
tags:
  - lora
  - qlora
  - chess
  - tinyllama
  - text-generation
  - adapter
---

# Model Card for `chezz-tinyllama-lora`

`chezz-tinyllama-lora` is a **LoRA adapter** that adds two skills to **TinyLlama-1.1B-Chat-v1.0**:

1. **Select a strong chess move** (matching a Stockfish-16 evaluation)  
2. **Explain that move and fire off a light-hearted taunt** — all emitted as a single JSON object.

Because only the rank-16 LoRA matrices are shared, the download is tiny (\< 100 MB) and license-clean: users still pull the original Apache-2.0 TinyLlama weights.

---

## Model Details

### Model Description

| | |
|---|---|
| **Developed by** | *Ducklin* (aka `ducklin404`) |
| **Financed by** | – |
| **Shared by** | *Ducklin* |
| **Model type** | LoRA adapter for a causal language model |
| **Languages** | English (natural language) + algebraic chess notation |
| **License** | Apache 2.0 |
| **Fine-tuned from** | `TinyLlama/TinyLlama-1.1B-Chat-v1.0` |

#### Model Sources

* **Code & weights:** <https://github.com/ducklin404/chezz>  
* **Dataset:** <https://huggingface.co/datasets/ducklin404/chezz_dataset>  
* **Dev-log / demo:** <https://asilentpond.com/projects/chezz>

---

## Uses

### Direct Use

* Drop-in chess-analysis chatbot that replies with:
  ```json
  {"from":"e2","to":"e4","piece":"P","explanation":"…","taunt":"…"}