Collections

Discover the best community collections!

Collections trending this week
Abliteration Weights( i made a better version above)
Here weight cuts caused random refusals on edge case prompts. I fixed this by adjusting the generation pipeline in the next version above
Qwen3.8-Flash-Next — Vector Quantized (MLX)
Data-free VQ rungs of Qwen3.8-Flash-Next for Apple Silicon. Four rungs, 2.1-5.5 bpw, MTP sidecar. Needs mlx-lm PR #1788. Built with VQLab.
Abliteration Weights( i made a better version above)
Here weight cuts caused random refusals on edge case prompts. I fixed this by adjusting the generation pipeline in the next version above
Qwen3.8-Flash-Next — Vector Quantized (MLX)
Data-free VQ rungs of Qwen3.8-Flash-Next for Apple Silicon. Four rungs, 2.1-5.5 bpw, MTP sidecar. Needs mlx-lm PR #1788. Built with VQLab.