Document Reranking on TREC DL (DL19 and DL20)
74.79NDCG@10 (DL19)Llama-3.1-8B Listwise
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Llama-3.1-8B ListwiseFine-tuning dataset=MS MARCO, Backbone=Llama-3.1-8B, Reranking Strategy=Listwise, Reranking depth=top 1002026.02 | 74.79 | 71.56 | |
| Qwen3-8B PointwiseFine-tuning dataset=MS MARCO, Backbone=Qwen3-8B, Reranking Strategy=Pointwise, Reranking depth=top 1002026.02 | 73.51 | 70.99 | |
| DiffuRank Perm-AssignFine-tuning dataset=MS MARCO, Backbone=LLaDA-1.5-8B, Reranking Strategy=Perm-Assign, Reranking depth=top 1002026.02 | 73.44 | 71.81 | |
| RankZephyr-7B ListwiseFine-tuning dataset=MS MARCO, Backbone=Zephyr-7B, Reranking Strategy=Listwise, Reranking depth=top 1002026.02 | 73.39 | 70.02 | |
| Qwen3-8B ListwiseFine-tuning dataset=MS MARCO, Backbone=Qwen3-8B, Reranking Strategy=Listwise, Reranking depth=top 1002026.02 | 73.15 | 70.75 | |
| DiffuRank PointwiseFine-tuning dataset=MS MARCO, Backbone=LLaDA-1.5-8B, Reranking Strategy=Pointwise, Reranking depth=top 1002026.02 | 73.06 | 69.69 | |
| DiffuRank Logits-ListFine-tuning dataset=MS MARCO, Backbone=LLaDA-1.5-8B, Reranking Strategy=Logits-List, Reranking depth=top 1002026.02 | 72.95 | 69.78 | |
| DiffuRank Perm-SampFine-tuning dataset=MS MARCO, Backbone=LLaDA-1.5-8B, Reranking Strategy=Perm-Samp, Sampling Steps (K)=4, Reranking depth=top 1002026.02 | 72.92 | 71.72 | |
| Llama-3.1-8B PointwiseFine-tuning dataset=MS MARCO, Backbone=Llama-3.1-8B, Reranking Strategy=Pointwise, Reranking depth=top 1002026.02 | 72.55 | 69.81 | |
| monoT5 (220M)Fine-tuning dataset=MS MARCO, Backbone=T5, Reranking depth=top 1002026.02 | 71.8 | 68.9 | |
| monoBERT (340M)Fine-tuning dataset=MS MARCO, Backbone=BERT, Reranking depth=top 1002026.02 | 70.5 | 67.28 | |
| LLaDA-1.5-8B ListwiseFine-tuning dataset=MS MARCO, Backbone=LLaDA-1.5-8B, Reranking Strategy=Listwise, Reranking depth=top 1002026.02 | 69.22 | 65.07 | |
| BM25Type=Unsupervised / Initial Retrieval2026.02 | 50.58 | 47.96 |