Tirth2109/qwen-indic-v1

qwen-indic-v1

LoRA fine-tune of Qwen/Qwen3-Embedding-8B on Indic-language data using a 3-stage contrastive recipe with teacher preservation. Last-token pooling, L2-normalized embeddings.

• —Base model: Qwen/Qwen3-Embedding-8B
• —Pooling: last-token
• —Languages: 22 scheduled Indic languages
• —Training data: train splits only; decontaminated against MTEB test sets