0xvoid0000/zira-researcher
LoRA-style (fewer layers) vs Qwen/Qwen3.5-4B. 0xvoid0000/zira-researcher matches part of Qwen/Qwen3.5-4B (426 of its 738 tensors; the rest of Qwen/Qwen3.5-4B has no counterpart). On the shared part: 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 32 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, MLP 96/96, Attention 32/224, Norms 0/105
2 derivatives checked
- mradermacher/zira-researcher-i1-GGUF: Quantization
- mradermacher/zira-researcher-GGUF: Same weights, converted