shanaka95/embeddinggemma-300m-rag-stage1

Update model card with Stage 1 eval results: R@1 0.175 → 0.393 (2.25x improvement over baseline)

Stage 1 fine-tune of embeddinggemma-300m on EnterpriseRAG-Bench: MNRL + Matryoshka dims [768,512,256,128], 8000 steps, bs=4, lr=2e-5, 200k unique training pairs. Eval R@1=0.45 (vs baseline 0.175 on 22q 5k corpus).

initial commit