DavidAU

/

Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODER

like 13

Hi David,

Thanks again for the release, and for your continued work for the community.

I had a few quick questions/requests:

1. Non-reasoning variant: Would it be feasible to train or release a faster “non-thinking” model (e.g., Gemma 3 27B–style) optimized for instruction following/roleplay without heavy reasoning?

2. Roleplay use case: In your experience, which models currently do best for empathetic roleplay? Lately many feel less warm; the older Lyra-4 family still seems stronger to me. If I’m configuring things wrong, I’d love any tips (settings, prompts, system messages).

3. Deployment docs: I usually serve via text-generation-webui / llama.cpp. The repo instructions look a bit out-of-date, could you revisit them or suggest a better deployment path?

Happy to share configs/logs if helpful. Also, if you have Patreon or Kickstarter, I’d be glad to support.

Thanks again