XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2
LoRA-style vs Qwen/Qwen3.5-0.8B-Base. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of k_proj, o_proj, q_proj, v_proj changed, in 6 of 24 layers; norms, biases and embeddings are untouched.
Tensors changed: Attention 24/220, Norms 0/135, Embeddings 0/5, MLP 0/127, other 0/1
1 derivatives checked
- mradermacher/ChunMengDie-0.8B-Qwen3.5-v1.2-GGUF: Same weights, converted