vwdubb/Swift-Qwen3.8-27B-Uncensored-MTP-Terse-Coder
Adapter vs Shockem/Qwen3.8-27b-Terse-Coder-LoRA. Shockem/Qwen3.8-27b-Terse-Coder-LoRA is a LoRA adapter, not full weights: rank 16, on down_proj, gate_proj, in_proj_a, in_proj_b, in_proj_qkv, in_proj_z, k_proj, o_proj… in 64 layers (496 adapted matrices). It modifies vwdubb/Swift-Qwen3.8-27B-Uncensored-MTP-Terse-Coder at load time.