uniem

Why we included this project

Chinese-language teams often struggle to find embedding models that match what English users take for granted. Uniem addresses that gap by shipping the M3E family of models on Hugging Face, and since they are fully sentence-transformers compatible, code that already loads a SentenceTransformer keeps working after you swap in the M3E name. The bundled FineTuner goes a step further, letting you fine-tune M3E, text2vec, or GPT-style checkpoints on your own data in a few lines when generic embeddings miss the mark on a domain-specific task. MTEB-zh, the included evaluation harness, runs classification and retrieval benchmarks across multiple Chinese datasets, so you can compare candidates on measured scores rather than gut feel. Teams building semantic search, RAG, or text classification for Chinese content are the obvious audience.