Benchmark for RAG
Using https://huggingface.co/datasets/nvidia/TechQA-RAG-Eval
| Item | Value |
|---|---|
| TopK | 10 |
| Embedding Model | text-embedding-3-small |
| Completion Model | google/gemini-2.5-flash |
| Evaluation Method | embedding |
| Documents | 496 |
| Queries | 50 |
| Sum Score | 2918.00 |
| Mean Score | 58.36 (🟠 Fair) |
| Hit@K | 0.9400 (🟡 Good) |
L2-normalization for embeddings
Minor retrieve improve.