Kodjaoglanian/Qwen-3.5-4B-ASVD-Healed
LoRA-style vs Qwen/Qwen3.5-4B. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of gate_proj, up_proj changed, in 32 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, MLP 64/64, Attention 0/192, Norms 0/105
1 derivatives checked
- Kodjaoglanian/Qwen-3.5-4B-ASVD-Healed-GGUF: Same weights, converted