Information Retrieval on NFCorpus (test)
0.462NDCG@10Ours
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| OursBackbone=GritLM, Training Protocol=supervised with relevance labels2026.02 | 0.462 | 58 | — | — | |
| Ours@30%Backbone=GritLM, Training Protocol=supervised with relevance labels, Fixed Dimension Ratio=30%2026.02 | 0.461 | 30 | — | — | |
| OursBackbone=OpenAI, Training Protocol=supervised with relevance labels2026.02 | 0.459 | 38 | — | — | |
| OursBackbone=Qwen-8B, Training Protocol=supervised with relevance labels2026.02 | 0.455 | 38 | — | — | |
| Ours@30%Backbone=OpenAI, Training Protocol=supervised with relevance labels, Fixed Dimension Ratio=30%2026.02 | 0.455 | 30 | — | — | |
| OursBackbone=Qwen-4B, Training Protocol=supervised with relevance labels2026.02 | 0.451 | 22 | — | — | |
| DIMEBackbone=OpenAI, Training Protocol=PRF-based2026.02 | 0.449 | 84 | — | — | |
| EclipseBackbone=OpenAI, Training Protocol=PRF-based2026.02 | 0.449 | 78 | — | — | |
| Ours@30%Backbone=Qwen-8B, Training Protocol=supervised with relevance labels, Fixed Dimension Ratio=30%2026.02 | 0.449 | 30 | — | — | |
| Ours@30%Backbone=Qwen-4B, Training Protocol=supervised with relevance labels, Fixed Dimension Ratio=30%2026.02 | 0.447 | 30 | — | — | |
| OursBackbone=L2V-LLaMA, Training Protocol=supervised with relevance labels2026.02 | 0.445 | 16 | — | — | |
| AdapterBackbone=L2V-LLaMA, Training Protocol=supervised with relevance labels2026.02 | 0.442 | 100 | — | — | |
| Ours@30%Backbone=L2V-LLaMA, Training Protocol=supervised with relevance labels, Fixed Dimension Ratio=30%2026.02 | 0.442 | 30 | — | — | |
| BaselineBackbone=OpenAI, Training Protocol=fully unsupervised2026.02 | 0.44 | 100 | — | — | |
| CutoffBackbone=OpenAI, Training Protocol=fully unsupervised2026.02 | 0.44 | 100 | — | — | |
| NormBackbone=OpenAI, Training Protocol=fully unsupervised2026.02 | 0.44 | 90 | — | — | |
| NormBackbone=L2V-LLaMA, Training Protocol=fully unsupervised2026.02 | 0.436 | 60 | — | — | |
| DIMEBackbone=Qwen-8B, Training Protocol=PRF-based2026.02 | 0.436 | 86 | — | — | |
| CutoffBackbone=Qwen-8B, Training Protocol=fully unsupervised2026.02 | 0.435 | 32 | — | — | |
| EclipseBackbone=Qwen-8B, Training Protocol=PRF-based2026.02 | 0.435 | 40 | — | — | |
| CutoffBackbone=L2V-LLaMA, Training Protocol=fully unsupervised2026.02 | 0.434 | 56 | — | — | |
| EclipseBackbone=L2V-LLaMA, Training Protocol=PRF-based2026.02 | 0.434 | 94 | — | — | |
| EclipseBackbone=GritLM, Training Protocol=PRF-based2026.02 | 0.434 | 94 | — | — | |
| AdapterBackbone=GritLM, Training Protocol=supervised with relevance labels2026.02 | 0.434 | 100 | — | — | |
| NormBackbone=Qwen-8B, Training Protocol=fully unsupervised2026.02 | 0.433 | 80 | — | — | |
| DIMEBackbone=L2V-LLaMA, Training Protocol=PRF-based2026.02 | 0.433 | 98 | — | — | |
| DIMEBackbone=GritLM, Training Protocol=PRF-based2026.02 | 0.433 | 24 | — | — | |
| BaselineBackbone=Qwen-8B, Training Protocol=fully unsupervised2026.02 | 0.432 | 100 | — | — | |
| BaselineBackbone=L2V-LLaMA, Training Protocol=fully unsupervised2026.02 | 0.432 | 100 | — | — | |
| CutoffBackbone=GritLM, Training Protocol=fully unsupervised2026.02 | 0.432 | 62 | — | — | |
| DIMEBackbone=Qwen-4B, Training Protocol=PRF-based2026.02 | 0.432 | 80 | — | — | |
| EclipseBackbone=Qwen-4B, Training Protocol=PRF-based2026.02 | 0.432 | 50 | — | — | |
| NormBackbone=GritLM, Training Protocol=fully unsupervised2026.02 | 0.43 | 92 | — | — | |
| BaselineBackbone=GritLM, Training Protocol=fully unsupervised2026.02 | 0.429 | 100 | — | — | |
| AdapterBackbone=OpenAI, Training Protocol=supervised with relevance labels2026.02 | 0.428 | 100 | — | — | |
| AdapterBackbone=Qwen-4B, Training Protocol=supervised with relevance labels2026.02 | 0.427 | 100 | — | — | |
| BaselineBackbone=Qwen-4B, Training Protocol=fully unsupervised2026.02 | 0.426 | 100 | — | — | |
| CutoffBackbone=Qwen-4B, Training Protocol=fully unsupervised2026.02 | 0.426 | 100 | — | — | |
| NormBackbone=Qwen-4B, Training Protocol=fully unsupervised2026.02 | 0.426 | 96 | — | — | |
| AdapterBackbone=Qwen-8B, Training Protocol=supervised with relevance labels2026.02 | 0.419 | 100 | — | — | |
| AdapterBackbone=L2V-Mistral, Training Protocol=supervised with relevance labels2026.02 | 0.417 | 100 | — | — | |
| EclipseBackbone=L2V-Mistral, Training Protocol=PRF-based2026.02 | 0.415 | 56 | — | — | |
| DIMEBackbone=L2V-Mistral, Training Protocol=PRF-based2026.02 | 0.414 | 50 | — | — | |
| OursBackbone=L2V-Mistral, Training Protocol=supervised with relevance labels2026.02 | 0.412 | 54 | — | — | |
| CutoffBackbone=L2V-Mistral, Training Protocol=fully unsupervised2026.02 | 0.411 | 62 | — | — | |
| NormBackbone=L2V-Mistral, Training Protocol=fully unsupervised2026.02 | 0.41 | 58 | — | — | |
| BaselineBackbone=L2V-Mistral, Training Protocol=fully unsupervised2026.02 | 0.407 | 100 | — | — | |
| Ours@30%Backbone=L2V-Mistral, Training Protocol=supervised with relevance labels, Fixed Dimension Ratio=30%2026.02 | 0.405 | 30 | — | — | |
| OursBackbone=Qwen-0.6B, Training Protocol=supervised with relevance labels2026.02 | 0.396 | 62 | — | — | |
| Ours@30%Backbone=Qwen-0.6B, Training Protocol=supervised with relevance labels, Fixed Dimension Ratio=30%2026.02 | 0.388 | 30 | — | — | |
| DIMEBackbone=Qwen-0.6B, Training Protocol=PRF-based2026.02 | 0.384 | 94 | — | — | |
| EclipseBackbone=Qwen-0.6B, Training Protocol=PRF-based2026.02 | 0.384 | 90 | — | — | |
| CutoffBackbone=Qwen-0.6B, Training Protocol=fully unsupervised2026.02 | 0.379 | 86 | — | — | |
| NormBackbone=Qwen-0.6B, Training Protocol=fully unsupervised2026.02 | 0.379 | 62 | — | — | |
| BaselineBackbone=Qwen-0.6B, Training Protocol=fully unsupervised2026.02 | 0.377 | 100 | — | — | |
| AdapterBackbone=Qwen-0.6B, Training Protocol=supervised with relevance labels2026.02 | 0.371 | 100 | — | — | |
| SPLADE++model=splade-cocondenser-ensembledistil, parameters=110 M, k1=1.2, b=0.752026.05 | 0.3495 | — | — | — | |
| Our ModelRetriever Type=Inference-free Sparse Retriever2024.11 | 0.349 | — | — | — | |
| SPLADE-v3-DistilRetriever Type=Sparse Retriever2024.11 | 0.348 | — | — | — | |
| SPLADE-doc-distillRetriever Type=Inference-free Sparse Retriever2024.11 | 0.34 | — | — | — | |
| SPLADE-v3-DocRetriever Type=Inference-free Sparse Retriever2024.11 | 0.338 | — | — | — | |
| ColBERTv2Retriever Type=Dense Retriever2024.11 | 0.338 | — | — | — | |
| SPLADE++-SelfDistilRetriever Type=Sparse Retriever2024.11 | 0.334 | — | — | — | |
| ContrieverRetriever Type=Dense Retriever2024.11 | 0.328 | — | — | — | |
| BM25Retriever Type=Inference-free Sparse Retriever2024.11 | 0.327 | — | — | — | |
| AgentIR BM252026.05 | 0.3267 | — | — | — | |
| TAS-BRetriever Type=Dense Retriever2024.11 | 0.319 | — | — | — | |
| TileMaxSimCorpus (dataset size)=3,633, Queries (number of queries)=323, Match=Exact, Embedding model=ColBERTv2, Embedding dimension=128-dim2026.06 | 0.31 | — | 0.533 | 26.3 | |
| Reference PyTorch MaxSimCorpus (dataset size)=3,633, Queries (number of queries)=323, Match=Exact, Embedding model=ColBERTv2, Embedding dimension=128-dim2026.06 | 0.31 | — | 0.533 | 26.3 |