clarkpalmer/GLM-5.3-Flash-Uncensored-MLX

GLM-5.3-Flash Uncensored MLX — Demo

Chat interface for orcarouter/GLM-5.3-Flash-Uncensored-MLX, a 320B-parameter abliterated vision-language model quantized for Apple Silicon.

Two modes

1. 1.Local MLX server — connect to your own mlx-vlm server running the uncensored weights. Supports vision (image + text).
2. 2.HF Inference Provider — try the base (censored) GLM-5.3-Flash via cloud API. Sign in with your HF account.

Run the model locally

pip install -U "mlx-vlm>=0.6.17"
python -m mlx_vlm.server --model orcarouter/GLM-5.3-Flash-Uncensored-MLX --port 8080

Then enter http://localhost:8080/v1 as the API base URL in the Space.