csqe

All results produced by QueryGym · fully reproducible!

12 model × retriever configurations for this method across BEIR, MS MARCO DL, and DL-HARD.
Click any row or the + button to expand. Tabs switch dataset context. The three steps (reformulate → retrieve → evaluate) update accordingly.

	Model	Retriever	ArguAna		DBPedia		FiQA		SciFact		COVID		News		BRIGHT — AOPS		BRIGHT — Biology		BRIGHT — Earth Science		BRIGHT — Economics		BRIGHT — LeetCode		BRIGHT — Pony		BRIGHT — Psychology		BRIGHT — Robotics		BRIGHT — Stack Overflow		BRIGHT — Sustainable Living		BRIGHT — TheoremQA Questions		BRIGHT — TheoremQA Theorems		DL-HARD		DL 2019		DL 2020
	Model	Retriever	nDCG@10	R@100	nDCG@10	R@100	nDCG@10	R@100	nDCG@10	R@100	nDCG@10	R@100	nDCG@10	R@100																									nDCG@10	R@1k	nDCG@10	R@1k	nDCG@10	R@1k
	Qwen2.5-72B-Instruct	BGE-base-en-v1.5	0.6229	0.9886	0.4024	0.4897	0.3796	0.7461	0.7484	0.9667	0.7793	0.1410	0.4626	0.4812	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.3757	0.8531	0.7179	0.8944	0.6687	0.8722
methodcsqe llmQwen2.5-72B-Instruct retrieverBGE-base-en-v1.5 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-arguana \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-arguana.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-arguana-test run.txt Run id f5fd7777f20b623f · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-dbpedia-entity \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-dbpedia-entity.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-dbpedia-entity-test run.txt Run id 94875e6bcad50a83 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-fiqa \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-fiqa.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-fiqa-test run.txt Run id 96a42cbd361934ea · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-scifact \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-scifact.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-scifact-test run.txt Run id 5a337650d6aec8fa · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-covid \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-covid.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-covid-test run.txt Run id 5a7dfe39f3fa4606 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-news \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-news.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-news-test run.txt Run id eeb5490fdadd7d2a · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.dlhard \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ /mnt/data/son/Thesis/t5/data/dlhard/neutral_queries.tsv run.txt Run id 38259981fc0cc7c5 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2019 \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl19-passage run.txt Run id 207f6f4e8b8b3331 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2020 \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl20-passage run.txt Run id 90f0e7cdd5df8856 · open full run detail →
	Qwen2.5-72B-Instruct	BM25	0.3864	—	0.3556	0.4639	0.2132	—	0.7141	—	0.6716	0.1491	0.3861	0.4892	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.2848	0.6998	0.6391	0.8608	0.5606	0.8603
methodcsqe llmQwen2.5-72B-Instruct retrieverBM25 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-arguana \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-arguana.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 \ beir-v1.0.0-arguana-test run.txt Run id 6e44c9a23591ec9a · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-dbpedia-entity \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-dbpedia-entity.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-dbpedia-entity-test run.txt Run id 43e3c6b01609e6f0 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-fiqa \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-fiqa.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 \ beir-v1.0.0-fiqa-test run.txt Run id be5818c6d67a2fc3 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-scifact \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-scifact.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 \ beir-v1.0.0-scifact-test run.txt Run id 07f8082534b1b82c · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-covid \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-covid.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-covid-test run.txt Run id a04f3be704b59a99 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-news \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-news.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-news-test run.txt Run id 7cc031ceec7f22cd · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.dlhard \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ /mnt/data/son/Thesis/t5/data/dlhard/neutral_queries.tsv run.txt Run id 374d868e8c9d9b43 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2019 \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl19-passage run.txt Run id 239707be9e6e9d0b · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2020 \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl20-passage run.txt Run id 4d261dda4cfb7038 · open full run detail →
	Qwen2.5-72B-Instruct	SPLADE++	0.5118	0.9787	0.3686	0.5021	0.3075	0.6521	0.6966	0.9433	0.6118	0.1082	0.3871	0.4548	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.2857	0.8246	0.6189	0.9070	0.5736	0.9052
methodcsqe llmQwen2.5-72B-Instruct retrieverSPLADE++ 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-arguana \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-arguana.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-arguana-test run.txt Run id 43c9e25451d58e11 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-dbpedia-entity \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-dbpedia-entity.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-dbpedia-entity-test run.txt Run id 49c7af4b43926798 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-fiqa \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-fiqa.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-fiqa-test run.txt Run id 868b26500a878476 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-scifact \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-scifact.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-scifact-test run.txt Run id dfae7d6c06e355d5 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-covid \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-covid.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-covid-test run.txt Run id 92ba76bc29a9d68f · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-news \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-news.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-news-test run.txt Run id 979e9554baa56930 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.dlhard \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ /mnt/data/son/Thesis/t5/data/dlhard/neutral_queries.tsv run.txt Run id c106bbf40f6cf8e6 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2019 \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl19-passage run.txt Run id 8057160502ca9c91 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2020 \ --method csqe \ --model Qwen/Qwen2.5-72B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl20-passage run.txt Run id f616925e2ae1ece5 · open full run detail →
	Qwen2.5-7B-Instruct	BGE-base-en-v1.5	0.6231	0.9893	0.3826	0.4879	0.3939	0.7437	0.7415	0.9727	0.7862	0.1449	0.4360	0.5126	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.3671	0.8348	0.7127	0.8803	0.6885	0.8850
methodcsqe llmQwen2.5-7B-Instruct retrieverBGE-base-en-v1.5 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-arguana \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-arguana.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-arguana-test run.txt Run id db84561b5bb02956 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-dbpedia-entity \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-dbpedia-entity.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-dbpedia-entity-test run.txt Run id da17c968549a5fc6 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-fiqa \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-fiqa.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-fiqa-test run.txt Run id 246fe58479a73dee · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-scifact \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-scifact.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-scifact-test run.txt Run id cebe92a4d886c671 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-covid \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-covid.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-covid-test run.txt Run id 1b3210148a98ea23 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-news \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-news.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-news-test run.txt Run id 7767b310f88ba350 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.dlhard \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ /mnt/data/son/Thesis/t5/data/dlhard/neutral_queries.tsv run.txt Run id ea85b70406c5f97f · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2019 \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl19-passage run.txt Run id 8a50e76e3972faef · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2020 \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl20-passage run.txt Run id 2ccc07bbc652dc9a · open full run detail →
	Qwen2.5-7B-Instruct	BM25	0.4008	0.9403	0.3767	0.5078	0.2200	0.5466	0.7183	0.9543	0.6757	0.1600	0.4504	0.5795	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.3322	0.7913	0.6873	0.8921	0.6083	0.8596
methodcsqe llmQwen2.5-7B-Instruct retrieverBM25 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-arguana \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-arguana.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-arguana-test run.txt Run id a4b96e532713cf8c · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-dbpedia-entity \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-dbpedia-entity.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-dbpedia-entity-test run.txt Run id 45f75d3dda634e3d · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-fiqa \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-fiqa.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-fiqa-test run.txt Run id 86d33a6522c7bf66 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-scifact \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-scifact.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-scifact-test run.txt Run id ed3f091219373725 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-covid \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-covid.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-covid-test run.txt Run id 46831c7ce6869053 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-news \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-news.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-news-test run.txt Run id 0966213446c0caca · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.dlhard \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ /mnt/data/son/Thesis/t5/data/dlhard/neutral_queries.tsv run.txt Run id a669e73a9f865bf9 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2019 \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"mode":"zs","num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl19-passage run.txt Run id 001ab6762f99c6b6 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2020 \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl20-passage run.txt Run id 2b82b2cea4817e13 · open full run detail →
	Qwen2.5-7B-Instruct	SPLADE++	0.5100	0.9801	0.3661	0.4830	0.3035	0.6521	0.6765	0.9527	0.6096	0.1024	0.4079	0.4866	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.3025	0.8057	0.6523	0.9089	0.6164	0.9039
methodcsqe llmQwen2.5-7B-Instruct retrieverSPLADE++ 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-arguana \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-arguana.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-arguana-test run.txt Run id e30beb7d6120032b · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-dbpedia-entity \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-dbpedia-entity.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-dbpedia-entity-test run.txt Run id c645b63d148df699 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-fiqa \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-fiqa.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-fiqa-test run.txt Run id 7c0e15a8a3ed05d7 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-scifact \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-scifact.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-scifact-test run.txt Run id d991a771a4ec16c3 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-covid \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-covid.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-covid-test run.txt Run id 6a5b94d82f568fbb · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-news \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-news.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-news-test run.txt Run id bfcb1e13cc699a61 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.dlhard \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ /mnt/data/son/Thesis/t5/data/dlhard/neutral_queries.tsv run.txt Run id 262563ce22f62374 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2019 \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl19-passage run.txt Run id 38aa528bf5037635 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2020 \ --method csqe \ --model Qwen/Qwen2.5-7B-Instruct \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl20-passage run.txt Run id 0947529629fd12ec · open full run detail →
	gpt-4.1	BGE-base-en-v1.5	0.6218	0.9915	0.4242	0.5229	0.4067	0.7384	0.7553	0.9633	0.7879	0.1431	0.4631	0.5075	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.4144	0.8640	0.7551	0.9009	0.7139	0.8968
methodcsqe llmgpt-4.1 retrieverBGE-base-en-v1.5 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-arguana \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-arguana.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-arguana-test run.txt Run id de46025c0b347808 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-dbpedia-entity \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-dbpedia-entity.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-dbpedia-entity-test run.txt Run id 9b434084e00ca34a · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-fiqa \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-fiqa.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-fiqa-test run.txt Run id d277bdcd328a483d · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-scifact \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-scifact.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-scifact-test run.txt Run id 7eb449e8dc5f1ad5 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-covid \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-covid.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-covid-test run.txt Run id 6c8f2c29a13e086a · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-news \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-news.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-news-test run.txt Run id 045d788a97b1e683 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.dlhard \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ /mnt/data/son/Thesis/t5/data/dlhard/neutral_queries.tsv run.txt Run id 6dcbed7ddbb1e73c · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2019 \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl19-passage run.txt Run id 4f79dd7d0fa93df0 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2020 \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl20-passage run.txt Run id 549fcfcdb38d82d8 · open full run detail →
	gpt-4.1	BM25	0.3977	0.9445	0.3899	0.5136	0.2473	0.5835	0.7206	0.9487	0.6994	0.1638	0.4790	0.5909	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.3658	0.7873	0.6899	0.9035	0.6548	0.8871
methodcsqe llmgpt-4.1 retrieverBM25 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-arguana \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-arguana.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-arguana-test run.txt Run id 2f7746328604d0bc · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-dbpedia-entity \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-dbpedia-entity.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-dbpedia-entity-test run.txt Run id 8b352f15c35044e7 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-fiqa \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-fiqa.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-fiqa-test run.txt Run id 52902b5453f351e6 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-scifact \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-scifact.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-scifact-test run.txt Run id b6b8168e2872eb9b · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-covid \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-covid.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-covid-test run.txt Run id 6fcfefc44e3d3a4c · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-news \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-news.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-news-test run.txt Run id d53af0d617c5b9be · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.dlhard \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ /mnt/data/son/Thesis/t5/data/dlhard/neutral_queries.tsv run.txt Run id ede9af17c87e4322 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2019 \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl19-passage run.txt Run id 0411fc9a522f86b1 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2020 \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl20-passage run.txt Run id a0441e8e51b3a613 · open full run detail →
	gpt-4.1	SPLADE++	0.3801	0.9829	0.3962	0.5232	0.3294	0.6748	0.7065	0.9593	0.6811	0.1116	0.4502	0.5018	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.3690	0.8341	0.6936	0.9193	0.6796	0.9397
methodcsqe llmgpt-4.1 retrieverSPLADE++ 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-arguana \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-arguana.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-arguana-test run.txt Run id 3e383ae339c3a62e · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-dbpedia-entity \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-dbpedia-entity.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-dbpedia-entity-test run.txt Run id b3f2134d9575311c · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-fiqa \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-fiqa.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-fiqa-test run.txt Run id f5ef9755a807ae98 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-scifact \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-scifact.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-scifact-test run.txt Run id b05b8be9006ec3a1 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-covid \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-covid.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-covid-test run.txt Run id 8be5974aec2d4609 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-news \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-news.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-news-test run.txt Run id 7017e4a12692cfa0 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.dlhard \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ /mnt/data/son/Thesis/t5/data/dlhard/neutral_queries.tsv run.txt Run id 322e4bb8fbbdbe9b · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2019 \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl19-passage run.txt Run id 6148ce92e9e11b2d · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2020 \ --method csqe \ --model openai/gpt-4.1 \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl20-passage run.txt Run id 435cc0da60103e45 · open full run detail →
	gpt-4.1-nano	BGE-base-en-v1.5	0.6210	0.9886	0.4147	0.5123	0.4112	0.7489	0.7583	0.9600	0.8174	0.1442	0.4351	0.4753	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.3516	0.8371	0.7304	0.8749	0.6873	0.8535
methodcsqe llmgpt-4.1-nano retrieverBGE-base-en-v1.5 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-arguana \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-arguana.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-arguana-test run.txt Run id 7ebe52879e43bf94 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-dbpedia-entity \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-dbpedia-entity.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-dbpedia-entity-test run.txt Run id 5ca04f4c42d9d563 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-fiqa \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-fiqa.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-fiqa-test run.txt Run id b26ff009f3ea9698 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-scifact \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-scifact.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-scifact-test run.txt Run id 20d45b2b113b7125 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-covid \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-covid.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-covid-test run.txt Run id 3f462b68dc170537 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-news \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-news.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-news-test run.txt Run id f13f81dff4e6cc16 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.dlhard \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ /mnt/data/son/Thesis/t5/data/dlhard/neutral_queries.tsv run.txt Run id 84dcf9caef28e7a7 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2019 \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl19-passage run.txt Run id 32492810415872de · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2020 \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BGE-base-en-v1.5 (dense) python -m pyserini.search.faiss \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.bge-base-en-v1.5 \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder BAAI/bge-base-en-v1.5 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl20-passage run.txt Run id 948ebd535661ce29 · open full run detail →
	gpt-4.1-nano	BM25	0.3964	0.9381	0.3647	0.4939	0.2401	0.5553	0.7099	0.9587	0.6171	0.1543	0.4271	0.5221	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.2436	0.7327	0.5410	0.8221	0.5142	0.8586
methodcsqe llmgpt-4.1-nano retrieverBM25 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-arguana \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-arguana.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-arguana-test run.txt Run id f32e376e0f39d46f · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-dbpedia-entity \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-dbpedia-entity.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-dbpedia-entity-test run.txt Run id bb094947e62ff61f · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-fiqa \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-fiqa.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-fiqa-test run.txt Run id 44d2d2aef166eb33 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-scifact \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-scifact.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-scifact-test run.txt Run id a7a782bb19e2ca65 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-covid \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-covid.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-covid-test run.txt Run id 46eb796f1598bbf8 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-news \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-news.flat \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-news-test run.txt Run id 4c83d57b9282b0b6 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.dlhard \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ /mnt/data/son/Thesis/t5/data/dlhard/neutral_queries.tsv run.txt Run id 6685bbc537e4cd4e · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2019 \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl19-passage run.txt Run id 4b8155969a7b03e2 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2020 \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · BM25 (lexical) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --bm25 --k1 0.9 --b 0.4 \ --output run.txt \ --hits 1000 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl20-passage run.txt Run id ce48ec0f783ddd1c · open full run detail →
	gpt-4.1-nano	SPLADE++	0.3792	0.9801	0.3805	0.5235	0.3256	0.6702	0.7055	0.9533	0.6313	0.1132	0.4193	0.4601	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.2789	0.7872	0.6134	0.8900	0.5883	0.9119
methodcsqe llmgpt-4.1-nano retrieverSPLADE++ 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-arguana \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-arguana.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-arguana-test run.txt Run id e0f4947fbccc6397 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-dbpedia-entity \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-dbpedia-entity.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-dbpedia-entity-test run.txt Run id 92ded4a648735d6b · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-fiqa \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-fiqa.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-fiqa-test run.txt Run id 481dd45b5c62d23c · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-scifact \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-scifact.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-scifact-test run.txt Run id 49cb755c774c5dbb · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-covid \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-covid.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-covid-test run.txt Run id f980832298d33f40 · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset beir-v1.0.0-trec-news \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index beir-v1.0.0-trec-news.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@100 python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.100 \ beir-v1.0.0-trec-news-test run.txt Run id b82e8af82f8b212c · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.dlhard \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ /mnt/data/son/Thesis/t5/data/dlhard/neutral_queries.tsv run.txt Run id cc25c04a032109ab · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2019 \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl19-passage run.txt Run id 7a7bacfd287897cd · open full run detail → 1 reformulate querygym → reformulated_queries.tsv python examples/querygym_pyserini/pipeline.py \ --dataset msmarco-v1-passage.trecdl2020 \ --method csqe \ --model openai/gpt-4.1-nano \ --steps reformulate \ --temperature 1 \ --max-tokens 128 \ --method-params '{"num_examples":4,"train_split":"train"}' \ --output-dir outputs/reproduce 2 retrieve pyserini · SPLADE++ (learned_sparse) python -m pyserini.search.lucene \ --threads 16 --batch-size 128 \ --index msmarco-v1-passage.splade-pp-ed \ --topics outputs/reproduce/queries/reformulated_queries.tsv \ --encoder naver/splade-cocondenser-ensembledistil \ --output run.txt \ --hits 1000 --impact 3 evaluate trec_eval · nDCG@10 + R@1k python -m pyserini.eval.trec_eval -c -m ndcg.cut.10 -m recall.1000 \ dl20-passage run.txt Run id e646c2b1b81107fa · open full run detail →