Large-scale 8B multilingual embedder delivering near-flagship quality at lower inference cost.
Access model weights, configuration files, and documentation.
See which devices can run this model and at what quality level.
The 8B variant of the F2LLM-v2 family from CodeFuse-AI, fine-tuned from Qwen3-8B-Base with the same fully open two-stage training recipe and 60M-sample multilingual corpus as its 14B sibling. Designed to deliver near-flagship quality at roughly half the inference footprint while preserving full reproducibility (weights, data, and code released).
View on Hugging Face