From the model card
Fields below are copied from the tags and counters on the HuggingFace repository unsloth/Qwen3-Embedding-4B at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.
- Publisher (HF namespace)
- unsloth
- Pipeline tag
- feature-extraction
- Library
- Sentence Transformers, Transformers
- Weight formats
- safetensors
- License tag
apache-2.0— read the license file in the repo before relying on it- Lineage
-
- base model Qwen/Qwen3-4B-Base
- fine-tune of Qwen/Qwen3-4B-Base
- Papers cited
- arXiv:2506.05176
- Downloads (HF counter at last fetch)
- 343,270
- Likes (HF counter at last fetch)
- 2
- Model card
- https://huggingface.co/unsloth/Qwen3-Embedding-4B
Use cases
- Semantic document retrieval in RAG pipelines
- Sentence similarity scoring
- Text-embeddings-inference production serving
- Embedding-based clustering of text corpora
- Benchmarking large embedding models on MTEB
Pros
- Apache 2.0 license
- 4B scale typically outperforms smaller embedding models on difficult retrieval tasks
- Sentence-transformers compatible for easy integration
- TEI-compatible for high-throughput serving
- Published MTEB results for objective comparison
Cons
- 4B is significantly larger than typical embedding models (0.3B-0.7B) — higher inference cost
- Unsloth conversion — not the official Qwen Embedding release
- Larger embedding models increase vector store memory at scale
- May require updated sentence-transformers version for Qwen3 architecture support
Tags
sentence-transformerssafetensorsqwen3text-generationtransformerssentence-similarityfeature-extractiontext-embeddings-inferencearxiv:2506.05176base_model:Qwen/Qwen3-4B-Basebase_model:finetune:Qwen/Qwen3-4B-Baselicense:apache-2.0endpoints_compatibleregion:us