Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model: unsloth/Qwen2.5-7B-Instruct
|
| 3 |
+
library_name: gguf
|
| 4 |
+
license: apache-2.0
|
| 5 |
+
tags: [gguf, tool-use, radicle, git, qwen2, qlora]
|
| 6 |
+
datasets: [h-d-h/rad-model-dataset]
|
| 7 |
+
---
|
| 8 |
+
|
| 9 |
+
# rad-model
|
| 10 |
+
|
| 11 |
+
QLoRA fine-tune of [Qwen 2.5 7B Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) for Radicle and Git tool calling. Distributed as a GGUF file for local inference with [llama.cpp](https://github.com/ggerganov/llama.cpp).
|
| 12 |
+
|
| 13 |
+
## Intended use
|
| 14 |
+
|
| 15 |
+
Tool-calling backend for CLI assistants that work with [Radicle](https://radicle.xyz) (decentralized code collaboration) and Git. The model selects and parameterizes the right CLI tool given a natural language request.
|
| 16 |
+
|
| 17 |
+
## Training
|
| 18 |
+
|
| 19 |
+
- **Method**: QLoRA (r=32, alpha=64) via [Unsloth](https://github.com/unslothai/unsloth)
|
| 20 |
+
- **Dataset**: [h-d-h/rad-model-dataset](https://huggingface.co/datasets/h-d-h/rad-model-dataset) — ~870 synthetic tool-calling examples covering 89 tools
|
| 21 |
+
- **Hardware**: NVIDIA RTX 3090 (24 GB VRAM)
|
| 22 |
+
- **Quantization**: Q4_K_M (~4.5 GB)
|
| 23 |
+
|
| 24 |
+
## Serving
|
| 25 |
+
|
| 26 |
+
```bash
|
| 27 |
+
llama-server -m rad-model-run6-q4_k_m.gguf --port 8080 -ngl 99 --host 0.0.0.0
|
| 28 |
+
```
|
| 29 |
+
|
| 30 |
+
The model serves an OpenAI-compatible `/v1/chat/completions` endpoint with tool-calling support.
|
| 31 |
+
|
| 32 |
+
## Evaluation
|
| 33 |
+
|
| 34 |
+
Evaluated on 88 held-out examples (stratified across all 89 tools). Scoring: 1.0 = correct tool + arguments, 0.75 = correct tool + extra args, 0.5 = correct tool + wrong args, 0.0 = wrong tool or no tool call.
|
| 35 |
+
|
| 36 |
+
See [RESULTS.md](https://app.radicle.xyz/nodes/rosa.radicle.xyz/rad:z2YCwgkXrZkUTu8c4CQayvk9Pkpky/tree/RESULTS.md) for full experiment history.
|
| 37 |
+
|
| 38 |
+
## Limitations
|
| 39 |
+
|
| 40 |
+
- Trained on synthetic data only; may not handle ambiguous real-world requests well
|
| 41 |
+
- Tool descriptions heavily influence accuracy — the base model with good descriptions can outperform the fine-tune on some tasks
|
| 42 |
+
- English only
|
| 43 |
+
- Designed for single-turn or short multi-turn tool-calling; not a general chat model
|
| 44 |
+
|
| 45 |
+
## Source
|
| 46 |
+
|
| 47 |
+
Developed on [Radicle](https://radicle.xyz): `rad:z2YCwgkXrZkUTu8c4CQayvk9Pkpky`
|
| 48 |
+
|
| 49 |
+
## License
|
| 50 |
+
|
| 51 |
+
Apache-2.0
|