joycx/Qwen3.8-27B-Velocity-GGUF
Adapter vs joycx/Qwen3.8-27B-Velocity. joycx/Qwen3.8-27B-Velocity is a LoRA adapter, not full weights: rank 32, on down_proj, gate_proj, in_proj_qkv, in_proj_z, k_proj, o_proj, out_proj, q_proj… in 64 layers (400 adapted matrices). It modifies joycx/Qwen3.8-27B-Velocity-GGUF at load time.