Hey everyone, made an uncensored version of Qwen3.5-4B - one of the brand new small models Qwen dropped these days.

Quick specs: 4B dense params, 32 layers, hybrid Gated DeltaNet linear attention + full softmax (3:1 ratio), 262K native context. Natively multimodal (text, image, video). This thing is surprisingly capable for its size .

This is the aggressive variant - 0/46