Textencoder gguf?

#1
by APCOTech - opened

Thank you so much for your efforts but can you quant the text encoder too?

You can use this INT4 version for now
https://huggingface.co/Abhiray/Minimax-H3-nvfp4-INT4-Convrot/blob/main/text_encoders/qwen3vl_32b_minimax_h3_int4_convrot.safetensors
It's roughly 15 GB, which is essentially the same size as a Q4_K_M quant would be, and it runs great even on consumer/older GPUs in ComfyUI. That said, I can still create a .gguf version of the text encoder soon if your workflow specifically needs it

It is for me and for many others who have GPUs with less vram. I actually downloaded this:
https://huggingface.co/Abiray/Minimax-H3-nvfp4-INT4-INT8-Convrot/blob/main/MiniMax_H3_FL2VA_pruned_int4_convrot.safetensors which is around 11GB. I asked since you are already doing the gguf quants so may you also do a quant for that huge text encoder.
but since we talked about it, which is better in quality? MiniMax_H3_FL2VA_pruned_int4_convrot.safetensors or MiniMax-H3-FL2VA-Q3_K_M.gguf?

MiniMax_H3_FL2VA_pruned_int4_convrot.safetensors always

Thank you a lot <3

Sign up or log in to comment