Retrieval on ArxivQA
81.2NDCG@5Qwen3-8B + DAPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen3-8B + DAPOCategory=DSL-R1, Backbone=Qwen3-8B, RL Strategy=DAPO2026.01 | 81.2 | 85 | |
| Qwen3-8B + GRPOCategory=DSL-R1, Backbone=Qwen3-8B, RL Strategy=GRPO2026.01 | 77.4 | 90 | |
| Qwen3-4B + DAPOCategory=DSL-R1, Backbone=Qwen3-4B, RL Strategy=DAPO2026.01 | 75.1 | 58 | |
| Qwen3-4B + GRPOCategory=DSL-R1, Backbone=Qwen3-4B, RL Strategy=GRPO2026.01 | 74.8 | 62 | |
| ColQwenCategory=Traditional Retrieval2026.01 | 73.9 | 42 | |
| GPT-4oCategory=LLM Based2026.01 | 70.5 | 35 | |
| Qwen3-8BCategory=LLM Based2026.01 | 61.2 | 155 | |
| Qwen3-4BCategory=LLM Based2026.01 | 55.4 | 105 | |
| BM25Category=Traditional Retrieval2026.01 | 31.6 | 25 |