Multi-source Answer Generation on CRAG Task 2 (test)
41Accuracy (%)DynaRAG
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DynaRAGUnderlying LLM=Llama 3-70B-Instruct, Evaluation Task=Task 2 (Web + API Context), API Fallback=Gorilla v2, Reranking=Enabled, Sufficiency Filtering=Enabled, Temperature=0, Top-p=0.92026.02 | 41 | 22.09 | 36.91 | |
| Direct RAGUnderlying LLM=Llama 3-70B-Instruct, Context Window Size=4K tokens, Reranking=Disabled, Sufficiency Checking=Disabled, Temperature=0, Top-p=0.92026.02 | 34.23 | 43.09 | 22.68 | |
| LLM OnlyUnderlying LLM=Llama 3-70B-Instruct, External Context=None, Temperature=0, Top-p=0.92026.02 | 28.53 | 34.95 | 36.52 |