Alibaba-NLP/gme-Qwen2-VL-7B-Instruct
LoRA-style vs Qwen/Qwen2-VL-7B-Instruct. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, k_proj, q_proj, up_proj, v_proj changed, in 28 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/2, Norms 0/187, MLP 84/216, Attention 84/324, lm_head 0/1
2 derivatives checked
- mradermacher/gme-Qwen2-VL-7B-Instruct-i1-GGUF: Quantization (fewer layers)
- mradermacher/gme-Qwen2-VL-7B-Instruct-GGUF: Same weights, converted (fewer layers)