barozp/Qwen3.8-27B-Opus-Distill-v2
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, k_proj, o_proj, q_proj, up_proj, v_proj changed, in 63 of 64 layers; norms, biases and embeddings are untouched.
Tensors changed: Norms 0/325, Attention 64/560, MLP 192/307, Embeddings 0/5, lm_head 0/1, other 0/1
7 derivatives checked
- barozp/Qwen3.8-27B-Opus-Distill-v2-GGUF: Same weights, converted
- barozp/Qwen3.8-27B-Opus-Distill-v2-FP8: Quantization
- barozp/Qwen3.8-27B-Opus-Distill-v2-MTPLX-4bit: Quantization
- barozp/Qwen3.8-27B-Opus-Distill-v2-MTPLX-3bit: Quantization
- barozp/Qwen3.8-27B-Opus-Distill-v2-MLX-4bit: Quantization
- caslca/Qwen3.8-27B-Opus-Distill-v2-mlx-uniform-4bit: Quantization
- barozp/Qwen3.8-27B-Opus-Distill-v2-MLX-8bit: Quantization