Error with mlx_lm (but not with LMstudio/mlx_lm!)
#1
by
balnazzar
- opened
First and foremost, thanks for this big DWQ!
I get the error you can see below while mlx_lm tries to download the model. Importantly, the model works perfectly with LM Studio with MLX as backend, beating the Unsloth 4-bit Dynamic Quant, inferenced with llama.cpp by a large margin (26 tok/s vs 18 tok/s).
Note the error is related to DNS resolution, but I never get any DNS error under any circumstances with any application whatsoever.
Importantly, I don't have any problem downloading and using any unsharded model with mlx_lm.
