Qwen/Qwen3-VL-Reranker-2B
LoRA-style vs Qwen/Qwen3-VL-2B-Instruct. 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 28 of 28 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/4, Norms 0/217, MLP 84/196, Attention 112/208
6 derivatives checked
- mradermacher/Qwen3-VL-Reranker-2B-i1-GGUF: Quantization (fewer layers)
- mradermacher/Qwen3-VL-Reranker-2B-GGUF: Same weights, converted (fewer layers)
- bealore/Qwen3-VL-Reranker-2B-GGUF: Same weights, converted (fewer layers)
- jeffpeng3/Qwen3-VL-Reranker-2B-NVFP4: Quantization
- nlpai-lab/RenderRank-2B: LoRA-style
- Surpem/Supertron2-Reranker-2B: Partial fine-tune
1 other repos hold byte-for-byte identical weight files.