model abliteration request.
hi.thanks for your great works.i tested qwen3coder and its performance are amazing.
please abliterate davidau coding models.
DavidAU/Qwen3-Coder-42B-A3B-Instruct-TOTAL-RECALL-MASTER-CODER-M
DavidAU/Qwen3-Esper3-Reasoning-CODER-Instruct-21B-Brainstorm20x-128k-ctx
The two models haven't received many likes. Is it necessary to proceed?
The two models haven't received many likes. Is it necessary to proceed?
I myself have only used DavidAU/Qwen3-53B-A3B-2507-TOTAL-RECALL-v2-MASTER-CODER. So cannot speak for the two he mentioned specifically.
However I have replaced claude sonnet 4 in my coding agents DavidAU/Qwen3-53B-A3B-2507-TOTAL-RECALL-v2-MASTER-CODER and DavidAU/Qwen3-42B-A3B-2507-Thinking-TOTAL-RECALL-v2-Medium-MASTER-CODER, couldn't be happier.
So to answer your question "Is it necessary?"; no, but if the compute isn't too expensive for you (time or money), why not :)
These models are really great for coding, but the limitations they have prevent them from showing their power. There's really no need to remove the limitations, but it would be great if you could, and I'd be grateful.
@huihui-ai
Quick note; If there is an abliterated version of the "core" model in any Brainstorm modified model, then I can use this version with Brainstorm,
and rebuild the model / upload ; in these cases "2507"s of Qwen models and "Coder" 30BA3B.
Likewise with Esper3, by Valiant labs.
Added note:
The 40x (and to some degree 20X) Brainstorm already de-censors (somewhat) the model.
This is not as strong as abliteration or fine tune(s) on a decensored dataset.
Update;
Source and quants up at :
I will be sure to like all of the models associated with huihui-ai. You're hands down the best thing on huggingface. I can shout you out on IG if you'd like - @officialmarionormil. DM me.
I will what I can do RE: Brainstorming
@huihui-ai
's models ; adding to the list.
Huihui ablit models are excellent.
I Love this model way too much
Update;
Source and quants up at :