Qwen 3.5 35B-A3B Uncensored Playground

A Jupyter notebook that runs Qwen3.5-35B-A3B-Uncensored (Q6_K, ~28 GB) locally via llama.cpp and serves a Gradio chat UI with streaming and adjustable parameters.

Requirements

• GPU: NVIDIA with CUDA 12.8+ and at least 32 GB VRAM (tested on A100 80GB)
• Python: 3.12
• Platform: Linux x86_64

Quick start with Modal (recommended)

The fastest way to get running — no local GPU needed:

1. Create a free account at modal.com
2. Go to Notebooks and upload qwen_uncensored.ipynb
3. In the sidebar, set the kernel GPU to A100 80GB (or any GPU with 32+ GB VRAM)
4. Run all cells — the model downloads, loads onto the GPU, and Gradio gives you a public chat URL

The whole setup takes about 5 minutes. You only pay for the seconds the kernel is running.

Quick start (local)

Open qwen_uncensored.ipynb and run all cells in order:

The Gradio UI provides a shareable public URL via share=True.

Default parameters

Defaults follow the official Qwen3.5 recommended settings for thinking mode (general tasks):

All parameters are adjustable via sliders in the sidebar.

Model

• Base: Qwen/Qwen3.5-35B-A3B — 35B total params, ~3B active per forward pass (MoE)
• Uncensored by: HauhauCS (Aggressive variant, 0/465 refusals)
• Quantization: Q6_K (6.58 BPW, imatrix)

License

MIT