Cross Lingual Semantic Embedding Generation

via “embedding generation for semantic similarity and retrieval”

text-generation model by undefined. 1,00,53,835 downloads.

Unique: Extracts embeddings from Qwen3-4B's final hidden layer (4096 dimensions), which are trained jointly with instruction-following objective, providing better semantic alignment for instruction-based queries than generic language models

vs others: More efficient than using separate embedding models like all-MiniLM-L6-v2 since inference is combined with generation; lower quality than specialized embedding models (e.g., BGE-large) but acceptable for many RAG applications; smaller embedding dimension than larger models reduces storage and comparison costs