bytkim/Qwen3.8-27B-pi
LoRA-style vs Qwen/Qwen3.8-27B. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, in_proj_a, in_proj_b, in_proj_qkv, in_proj_z, k_proj, o_proj, out_proj, q_proj, up_proj, v_proj changed, in 64 of 64 layers; norms, biases and embeddings are untouched.
Tensors changed: Norms 0/325, Attention 304/560, MLP 192/307, Embeddings 0/5, lm_head 0/1, other 0/1
6 derivatives checked
- bartowski/bytkim_Qwen3.8-27B-pi-GGUF: Same weights, converted
- leonsarmiento/Qwen3.8-27B-pi-4bit-mtp-mlx: Quantization
- firetussin/Qwen3.8-27B-pi-RCO-allocation-transfer-GGUF: Quantization
- maeuryhoisj/Qwen3.8-27B-pi-exl3-5bpw: Same weights, converted
- grandslam77/Qwen3.8-27B-pi-NVFP4: Quantization
- mrq/Qwen3.8-27B-pi-exl3-3.00bpw: Same weights, converted