Qwen/Qwen3-VL-Embedding-8B
LoRA-style vs Qwen/Qwen3-VL-8B-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 36 of 36 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/4, Norms 0/261, MLP 108/232, Attention 108/252
12 derivatives checked
- RamManavalan/Qwen3-VL-Embedding-8B-FP8: Quantization
- VesNFF/Qwen3-VL-Embedding-8B-GGUF: Same weights, converted (fewer layers)
- mradermacher/Qwen3-VL-Embedding-8B-GGUF: Same weights, converted (fewer layers)
- mradermacher/Qwen3-VL-Embedding-8B-i1-GGUF: Quantization (fewer layers)
- QuaduxIT/Qwen3-VL-Embedding-8B-W8A16: Quantization
- batiai/Qwen3-VL-Embedding-8B-GGUF: Quantization (fewer layers)
- dam2452/Qwen3-VL-Embedding-8B-GGUF: Same weights, converted (fewer layers)
- collin-park/Qwen3-VL-Embedding-8B-W8A8: Quantization
- gonuit/Qwen3-VL-Embedding-8B-AWQ-4bit: Quantization
- jedisct1/Qwen3-VL-Embedding-8B-mlx: Quantization
- Zeknes/Qwen3-VL-Embedding-8B-MLX-4bit: Quantization
- nkamiy/Qwen3-VL-Embedding-8B-8bit-mlx: Quantization
5 other repos hold byte-for-byte identical weight files.