Qwen/Qwen3-Embedding-4B
LoRA-style vs Qwen/Qwen3-4B-Base. 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 36 of 36 layers; norms, biases and embeddings are untouched.
Tensors changed: Norms 0/145, MLP 108/108, Attention 144/144
37 derivatives checked
- boboliu/Qwen3-Embedding-4B-W4A16-G128: Quantization
- mlx-community/Qwen3-Embedding-4B-4bit-DWQ: Quantization
- furiosa-ai/Qwen3-Embedding-4B: Identical copy
- Octen/Octen-Embedding-4B: Full fine-tune
- Octen/Octen-Embedding-4B-INT8: Quantization
- mradermacher/Qwen3-Embedding-4B-i1-GGUF: Quantization
- drawais/Qwen3-Embedding-4B-AWQ-INT4: Quantization
- Querit/Querit-4B: Full fine-tune
- mradermacher/Qwen3-Embedding-4B-GGUF: Same weights, converted
- ufca-llms/jua-4B-mixed: LoRA-style
- QuaduxIT/Qwen3-Embedding-4B-W8A16: Quantization
- AQ-MedAI/Diver-Retriever-4B: Full fine-tune
- itsbohara/qwen3_embedding_4b_q8_0: Full fine-tune
- Casual-Autopsy/Qwen3-Embedding-4B-GGUFs: Same weights, converted
- yale-nlp/RTriever-4B: LoRA-style
- bflhc/MoD-Embedding: LoRA-style
- Hanno-Labs/dinghy-law-4b-v1: LoRA-style
- AQ-MedAI/Diver-Retriever-4B-1020: Full fine-tune
- cstr/qwen3-embed-4b-GGUF: Quantization
- Tevatron/AgentIR-4B: LoRA-style
- enacimie/Qwen3-Embedding-4B-Q4_K_M-GGUF: Quantization
- DCC-BS/Qwen3-Embedding-4B-FP8-Dynamic: Quantization
- HIT-TMG/JevEmbed-Qwen3-Embedding-4B: LoRA-style
- lainlives/Qwen3-Embedding-4B-bnb-4bit: Quantization
- RKNNAI/RK182X-EMBEDDING-Qwen3-Embedding-4B: Quantization
- fabricioalmeida/BumbaLM-Embedding-4B-v0.1: Same weights, converted
- batiai/Qwen3-Embedding-4B-GGUF: Quantization
- ironbcc/Qwen3-Embedding-4B-NVFP4: Quantization
- novacardsai/qwen3-medical: Full fine-tune
- dolfsai/Qwen3-Embedding-4B-W4A16_ASYM: Quantization
18 other repos hold byte-for-byte identical weight files.