Question Answering on MMLU Med
90.6AccuracyHCQR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HCQRLLM Backbone=Qwen3 30B, Context Budget=15 documents2026.03 | 90.6 | — | |
| CoTLLM Backbone=Qwen3 30B, Context Budget=15 documents2026.03 | 90.2 | — | |
| Rerank-RAGLLM Backbone=Qwen3 30B, Context Budget=15 documents2026.03 | 90.1 | — | |
| HyDELLM Backbone=Qwen3 30B, Context Budget=15 documents2026.03 | 89.4 | — | |
| Simple RAGLLM Backbone=Qwen3 30B, Context Budget=15 documents2026.03 | 88.9 | — | |
| RewritingLLM Backbone=Qwen3 30B, Context Budget=15 documents2026.03 | 88.8 | — | |
| MAIN-RAGLLM Backbone=Qwen3 30B, Context Budget=15 documents2026.03 | 88.2 | — | |
| HCQRLLM Backbone=Qwen3 4B, Context Budget=15 documents2026.03 | 85.4 | — | |
| CoTLLM Backbone=Qwen3 4B, Context Budget=15 documents2026.03 | 82.9 | — | |
| HyDELLM Backbone=Qwen3 4B, Context Budget=15 documents2026.03 | 82.6 | — | |
| RewritingLLM Backbone=Qwen3 4B, Context Budget=15 documents2026.03 | 82.5 | — | |
| Rerank-RAGLLM Backbone=Qwen3 4B, Context Budget=15 documents2026.03 | 82.3 | — | |
| Simple RAGLLM Backbone=Qwen3 4B, Context Budget=15 documents2026.03 | 81.8 | — | |
| MAIN-RAGLLM Backbone=Qwen3 4B, Context Budget=15 documents2026.03 | 81.2 | — | |
| HCQRLLM Backbone=Llama3.1 8B, Context Budget=15 documents2026.03 | 79.7 | — | |
| HyDELLM Backbone=Llama3.1 8B, Context Budget=15 documents2026.03 | 76.5 | — | |
| Rerank-RAGLLM Backbone=Llama3.1 8B, Context Budget=15 documents2026.03 | 75.8 | — | |
| MAIN-RAGLLM Backbone=Llama3.1 8B, Context Budget=15 documents2026.03 | 75.7 | — | |
| CoTLLM Backbone=Llama3.1 8B, Context Budget=15 documents2026.03 | 75 | — | |
| Simple RAGLLM Backbone=Llama3.1 8B, Context Budget=15 documents2026.03 | 74.7 | — | |
| RewritingLLM Backbone=Llama3.1 8B, Context Budget=15 documents2026.03 | 74.7 | — | |
| FullBackbone=Llama-3.1-8B, Training Data Percentage=100%, Data Selection Strategy=All data2026.04 | 71.33 | — | |
| HCQRLLM Backbone=Llama3.2 3B, Context Budget=15 documents2026.03 | 68.6 | — | |
| HyDELLM Backbone=Llama3.2 3B, Context Budget=15 documents2026.03 | 67.8 | — | |
| Rerank-RAGLLM Backbone=Llama3.2 3B, Context Budget=15 documents2026.03 | 67.1 | — | |
| RewritingLLM Backbone=Llama3.2 3B, Context Budget=15 documents2026.03 | 67 | — | |
| MAIN-RAGLLM Backbone=Llama3.2 3B, Context Budget=15 documents2026.03 | 67 | — | |
| CoTLLM Backbone=Llama3.2 3B, Context Budget=15 documents2026.03 | 66.3 | — | |
| wICIBackbone=Llama-3.1-8B, Training Data Percentage=30%, Data Selection Strategy=In-Context Influence2026.04 | 65.33 | — | |
| Simple RAGLLM Backbone=Llama3.2 3B, Context Budget=15 documents2026.03 | 64.5 | — | |
| FullBackbone=Mistral-7B-v0.3, Training Data Percentage=100%, Data Selection Strategy=All data2026.04 | 51.67 | — | |
| wICIBackbone=Mistral-7B-v0.3, Training Data Percentage=30%, Data Selection Strategy=In-Context Influence2026.04 | 50 | — | |
| RandomBackbone=Llama-3.1-8B, Training Data Percentage=30%, Data Selection Strategy=Random sampling2026.04 | 45.67 | — | |
| RandomBackbone=Mistral-7B-v0.3, Training Data Percentage=30%, Data Selection Strategy=Random sampling2026.04 | 36 | — | |
| CoTLLM Backbone=Average, Context Budget=15 documents2026.03 | — | 78.6 | |
| HCQRLLM Backbone=Average, Context Budget=15 documents2026.03 | — | 81.1 | |
| HyDELLM Backbone=Average, Context Budget=15 documents2026.03 | — | 79.1 | |
| MAIN-RAGLLM Backbone=Average, Context Budget=15 documents2026.03 | — | 78 | |
| Rerank-RAGLLM Backbone=Average, Context Budget=15 documents2026.03 | — | 78.8 | |
| RewritingLLM Backbone=Average, Context Budget=15 documents2026.03 | — | 78.2 | |
| Simple RAGLLM Backbone=Average, Context Budget=15 documents2026.03 | — | 77.5 |