README.md

---
title: My Pi Agent
emoji: 🦀
colorFrom: gray
colorTo: indigo
sdk: gradio
sdk_version: 6.14.0
python_version: '3.12'
app_file: app.py
pinned: false
models:
  - AlexWortega/qwen35-4b-soyuz-merged-gguf
---

# My Pi Agent — Soyuz

Chat with **qwen35-4b-soyuz-merged** (a Qwen3.5-4B hybrid linear-attention model)
served as a GGUF via `llama-cpp-python` on **ZeroGPU**.

The right-hand panel surfaces what the model actually produced:
- **🧠 Reasoning** — the `<think> ... </think>` chain of thought.
- **📝 Raw output** — the full untouched generation, tags included.

GGUF: [`AlexWortega/qwen35-4b-soyuz-merged-gguf`](https://huggingface.co/AlexWortega/qwen35-4b-soyuz-merged-gguf)
(`Q4_K_M`, MTP head dropped — `--no-mtp` — since llama.cpp does not yet run Qwen3.5 MTP).