If possible, please make GGUF Q8 version too

#3
by MavenDE - opened

Something to hit 5090 32GB VRAM + 256GB RAM limit

This one seems to be uncensored. Please release Q8 if you are able to and then fixed fp16 470GB.

Is Q4 so bad? Also why not Q6 (it should practically be almost lossless).

I tested these models carefully. Everything worse than Q8 is miserable and not worth 1 minute of time.
FP16 and Q8 only makes sense. Q4 is a joke and waste of electricity and time

We will release Q8, but you need to pay for it.

199$ or more? :)

199$ or more? :)

Yes, 199$ will do.

hf.co currently offers 1TB of private storage space for a monthly fee of $25. Shared storage space also has limitations now, and files cannot be uploaded freely.

I tested these models carefully. Everything worse than Q8 is miserable and not worth 1 minute of time.
FP16 and Q8 only makes sense. Q4 is a joke and waste of electricity and time

Interesting. Did you test Unsloth UD quants by chance?

199$ or more? :)

Yes, 199$ will do.

hf.co currently offers 1TB of private storage space for a monthly fee of $25. Shared storage space also has limitations now, and files cannot be uploaded freely.

Mediascope then? Or through torrent sharing?

And "thinking" version is more useful. It will be funny to read its abliterated thoughts.

199$ or more? :)

Yes, 199$ will do.

hf.co currently offers 1TB of private storage space for a monthly fee of $25. Shared storage space also has limitations now, and files cannot be uploaded freely.

Mediascope then? Or through torrent sharing?

Download using the token from hf.co.
huggingface-cli download huihui-ai/Huihui-Qwen3-VL-235B-A22B-Instruct-abliterated-Q8_0-GGUF --local-dir ./huihui-ai/Huihui-Qwen3-VL-235B-A22B-Instruct-abliterated-Q8_0-GGUF --token xxx

I made abliterated model too, if someone is interested.
https://huggingface.co/Nekotekina/Qwen3-VL-235B-A22B-Projected-Abliterated-GGUF

Life saver from money makers! Is there instruct or thinking one?

It's Instruct, forgot to type the name correctly. I might try to make Thinking later although I don't generally like Thinking models.

Sign up or log in to comment