---
license: apache-2.0
tags:
  - tarx
  - local-ai
  - gguf
  - llama-cpp
---
# TARX Qwen2.5 7B Deep (Q4_K_M)

Fine-tuned reasoning model for TARX Workbench. Local-first AI inference.

- **Base:** Qwen 2.5 7B
- **Fine-tune:** 2-round LoRA (502 examples, loss 1.237)
- **Quantization:** Q4_K_M (4.4GB)
- **Speed:** ~18 tok/s on Apple M4
- **Context:** 4096 tokens

## Usage with llama-server
```bash
llama-server --hf-repo tarx-ai/tarx-qwen2.5-7b-deep:Q4_K_M --port 11435
```

## Usage with TARX CLI
```bash
curl -fsSL https://tarx.com/install | sh
tarx start
tarx chat "hello"
```

Local. Private. Proactive.