Pith. sign in

REVIEW 3 major objections 6 minor 31 references

An Alternative to FLOPS Regularization to Effectively Productionize SPLADE-Doc

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read DF-FLOPS, a document-frequency-weighted variant of FLOPS regularization, cuts SPLADE-Doc's production retrieval latency roughly tenfold while preserving effectiveness and improving out-of-domain retrieval on 12 of 13 tasks.

desk verdict A simple, effective regularizer that cuts SPLADE-Doc latency by an order of magnitude; the core idea is new and the results are strong, though reproducibility gaps and the unablated DF-estimation schedule need attention. read the letter →

arxiv 2505.15070 v1 pith:QIRGPJBH submitted 2025-05-21 cs.IR cs.CL

classification cs.IRcs.CL
keywords learnedsparseretrievalSPLADEFLOPSregularizationdocumentfrequencylatencyinvertedindexproductionsearchzero-shot
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Learned sparse retrieval models such as SPLADE-Doc are fast in principle but slow in practice because FLOPS regularization limits how many terms each document vector uses, not how many documents each term appears in; a single stopword can end up in over 95% of the collection and force the search engine to score almost every passage. This paper proposes DF-FLOPS, a training-time regularizer that weights each term's FLOPS penalty by a nonlinear function of its document frequency, so the model is penalized for overusing globally frequent tokens and learns to reserve them for passages where they are genuinely informative. On the MS MARCO corpus, DF-FLOPS reduces the most frequent token's document frequency from 95.8% to 8.0%, cuts average retrieval latency from 922 ms to 161 ms (and to 88 ms with post-hoc pruning@150) on a production search engine, and keeps MRR@10 within 2.2 points of the FLOPS-trained model while improving Recall@1000. In zero-shot evaluation on BEIR, DF-FLOPS outperforms FLOPS on 12 of 13 datasets. The paper's conclusion is that term-level sparsity, not merely document-level sparsity, is what makes learned sparse retrieval viable for production deployment.

What carries the argument

The load-bearing object is the DF-FLOPS regularizer: a per-term reweighting of the FLOPS loss in which each vocabulary term $t$ contributes $(w_t/N \sum_i r_{i,t})^2$ and the weight $w_t$ is a generalized logistic function of the term's document-frequency ratio $DF_t/|C|$. With cutoff $\alpha$ and steepness $\beta$, terms whose document frequency exceeds the cutoff are penalized much more heavily, shifting the model's inductive bias from producing many low-weight tokens everywhere to concentrating weight on content terms. The machinery also includes the periodic DF estimation procedure, which recomputes $DF_t$ from a 50,000-pair validation sample every 100 training steps and keeps the penalty aligned with the model's current behavior without extra GPU cost. Because the query encoder is a binary bag of tokens, all of the efficiency burden sits on document posting lists, which is exactly what the DF weighting shortens.

What would settle it

Train DF-FLOPS on a corpus whose term distribution shifts over time, such as a multi-year news stream, using the same 50,000-pair validation sample refreshed every 100 steps; if the most frequent token's document frequency rises back toward the FLOPS level and average latency regresses toward 900 ms, the periodic estimation is too stale for production use.

Watch

Extended reading notes

Core claim

The central claim is that the failure of FLOPS-trained SPLADE-Doc in production is a term-frequency problem, not a per-vector density problem, and that penalizing high document frequencies directly fixes it. The paper defines the DF-FLOPS loss as $$\ell_{DF-FLOPS}=\sum_{t\in V}\left(\frac{w_t}{N}\sum_{i=1}^{N} r_{i,t}\right)^2,\quad w_t=\mathrm{activ}(DF_t/|C|),$$ which reduces to ordinary FLOPS when $w_t=1$. The activation is a generalized logistic function with a frequency cutoff $\alpha=0.1$ and steepness $\beta=10$, so terms appearing in more than about 10% of the corpus are strongly penalized. Document frequencies are estimated every 100 training steps from a 50,000-pair validation sample, since exact corpus-wide DFs are too expensive to compute online. The result is that SPLADE-Doc with DF-FLOPS retains semantically salient high-frequency terms (e.g., 'who' for the World Health Organization) while dropping stopwords that FLOPS keeps, shortening posting lists from an average of 8.6 million matches per query to 1.9 million, and bringing average latency near BM25's.

Load-bearing premise

The load-bearing premise is that document frequencies estimated from a 50,000-pair sample every 100 training steps stay close to the true corpus-wide frequencies; the static version computed once before training performed much worse, so the periodic update is essential.

Editorial extensions

If this is right

  • If DF-FLOPS is adopted, SPLADE-Doc can serve MS MARCO queries at an average latency of 161 ms, or 88 ms with pruning@150, versus 922 ms for the FLOPS-trained model, bringing learned sparse retrieval within about 1.3x of BM25's 69 ms.
  • The most frequent token's document frequency falls from 95.8% to 8.0%, and average matches per query fall from 8.6 million to 1.9 million, so the inverted index performs far less work per query.
  • Compared with the original FLOPS model, DF-FLOPS drops MRR@10 by 2.2 points (32.2 to 30.0) while Recall@1000 edges up from 92.4 to 92.9; compared with the equally fast high-regularization FLOPS model, it gains 1.7 MRR@10 and 4.5 Recall@1000.
  • Out-of-domain, DF-FLOPS improves nDCG@10 over the FLOPS baseline on 12 of 13 BEIR datasets, with especially large gains on quora (7.6 to 48.1) and arguana (11.2 to 33.3).
  • Because the penalty is applied during training, DF-FLOPS composes with inference-time top-k pruning: adding pruning@150 lowers average latency from 161 ms to 88 ms with almost no effectiveness change (MRR@10 29.7 vs 30.0).

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same per-term DF weighting could be dropped into other learned sparse architectures (e.g., DeepImpact, EPIC, or UniCOIL) that train with a sum-over-terms loss; if the inductive-bias story holds, they should see comparable posting-list shrinkage and latency gains.
  • An online running average of document frequencies would remove the dependence on a fixed validation sample and could make DF-FLOPS reliable when the corpus distribution drifts during training.
  • The cross-domain gains hint that DF-FLOPS regularizes against overfitting to dataset-specific frequent tokens; a direct test would be training on a small or noisy corpus to see whether the out-of-domain margin over FLOPS grows.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes DF-FLOPS, a variant of FLOPS regularization for SPLADE-Doc that scales the per-term penalty by an activation of the term's document frequency, thereby discouraging high-DF terms and shortening posting lists. Document frequencies are periodically re-estimated during training from a 50K validation sample every 100 steps. Experiments on MS-Marco, TREC DL 2019/2020, and BEIR, with latency measured in Apache Solr, show that DF-FLOPS reduces the top-token DF from 95.8% to 8.0%, cuts average latency from 922 ms to 161 ms (87.8 ms with pruning@150), and reports improved effectiveness on 12 of 13 BEIR datasets relative to FLOPS.

Significance. If substantiated, the contribution is practically significant: it provides a training-time regularizer that brings SPLADE-Doc retrieval latency close to BM25 in a full-featured production engine, without bespoke indexing or query-time pruning. The use of a production Solr engine, the reporting of number of matches and P99 latencies, and the comparison against stronger FLOPS regularization and static pruning are strengths. However, the main empirical claims rest on single runs and on a periodic DF schedule that is never ablated, and the paper does not disentangle the DF weighting from the re-estimation schedule. These issues need to be addressed before the contribution is fully established.

major comments (3)
  1. [§2 'Approximating DF' and §4 'RQ4'] The successful DF-FLOPS configuration periodically re-estimates document frequencies from a 50K validation sample every 100 training steps, while the static variant that precomputes DFs before training performs considerably worse (MRR@10=28.8, Top@1 Token DF=51.7%). Because neither the update interval nor the sample size is varied, the observed gains could be due to the periodic refresh of the penalty targets rather than to the DF weighting itself. This is load-bearing for the central claim that DF-FLOPS, as opposed to its training schedule, productionizes SPLADE-Doc. Please add ablations that (a) fix DFs computed from a converged FLOPS model, (b) vary the update interval (e.g., 50, 200, 500 steps), and (c) vary the estimation sample size, and report Top@1 DF and Solr latency for each configuration.
  2. [§4, Tables 2 and 3] All effectiveness numbers appear to come from a single training run, with no error bars or significance tests. The key in-domain comparisons (MRR@10 32.2 vs. 30.0; 29.2 vs. 30.0) are small, and the BEIR '12 of 13' claim aggregates per-dataset differences of very different magnitudes. Please report multiple seeds or at least paired bootstrap significance tests on MS-Marco and BEIR, and confidence intervals for the latency measurements, which are currently averaged over only three repetitions.
  3. [§2–§3 (Eq. 2) and §4 (RQ3/RQ4)] FLOPS and DF-FLOPS are compared using different lambda ranges (FLOPS up to 1, DF-FLOPS up to 1e3), justified by the fact that the DF-FLOPS loss is smaller since w_t <= 1. The paper does not report the effective penalty magnitude (e.g., mean w_t or the product lambda times the loss) during training, so it is unclear whether the improvement comes from the term weighting itself or simply from re-scaling the regularizer. Please report the effective regularization magnitude and include a lambda sweep for DF-FLOPS to show that the result is not driven by the choice of lambda.
minor comments (6)
  1. [§2, Eq. (3)] Equation (3) is ambiguous as typeset; please define the exponent explicitly (e.g., x^{log_alpha 2} - 1 versus x log_alpha 2 - 1).
  2. [§1 and §4] Table 1 is referenced as 'Table ??' in Sections 1 and 4; fix the cross-references.
  3. [References] References [4] and [5] both cite the TREC 2019 overview; one of them should be the TREC 2020 overview.
  4. [Table 3] The asterisk on BM25 values in Table 3 should be explained in the caption, identifying which values are copied from [8] or computed with PISA, as is done in Table 2.
  5. [§4, RQ4] The sentence 'DF-FLOPS significantly decreases the latency' should be replaced with the concrete numbers (e.g., 95.8% to 8.0% for Top@1 DF), since no significance test is reported.
  6. [§3] A reproducibility statement covering code/model release and the exact Solr configuration (caching, warm-up, concurrency) would strengthen the production-latency claims.

Circularity Check

1 steps flagged · score 3.0 of 10

DF-FLOPS's reduction of high-DF terms is encoded in its loss; the latency and zero-shot results are independent, but the static-variant gap remains unablated.

  1. self definitional [Section 2, Eq. (2); Section 4, RQ4]
    "DF-FLOPS regularization aims to address this limitation by scaling the loss of each term by a weight 𝑤𝑡 depending on the term’s document frequency. ... ℓ𝐷𝐹−𝐹𝐿𝑂𝑃𝑆 = ∑𝑡∈𝑉 (𝑤𝑡/𝑁 ∑𝑁𝑖=1 𝑟𝑖,𝑡)^2, where 𝑤𝑡 =𝑎𝑐𝑡𝑖𝑣(𝐷𝐹𝑡/|𝐶|). ... The DF-FLOPS significantly decreases the latency by lowering the DF% of high frequent tokens (down to 8% for the most frequent token)."

    The per-term penalty weight w_t is a monotone increasing function of DF_t, so minimizing this loss directly drives down the weights of high-DF terms. Reporting that DF-FLOPS reduces Top@1 Token DF (95.8% to 8.0%) is therefore a restatement of the optimization objective, not an emergent empirical prediction. The independent content of the paper lies in the measured Solr latency reduction and BEIR zero-shot effectiveness, which are not entailed by the loss definition alone.

full rationale

The paper's only circular element is the claim that DF-FLOPS reduces high-DF terms: Eq. (2) defines the regularizer to penalize terms according to a monotone function of their document frequency, so the observed DF reduction is a direct consequence of the loss rather than an independent finding. The central productionization claim is nevertheless supported by external evidence that is not forced by construction: actual Solr retrieval latency drops from 922.0 ms to 161.0 ms, and DF-FLOPS improves over FLOPS on 12 of 13 BEIR tasks. There is no load-bearing self-citation chain; the one overlapping-author reference (PISA bindings for a BM25 baseline) is a computational tool, not an argumentative premise. The manuscript's own limitations—the fixed 100-step and 50K-sample DF re-estimation schedule, the unablated static-variant failure, and hyperparameters α, β tuned on the dev set—are real robustness concerns but are not circularity. Overall, the core contribution remains independently testable, so the score is modest.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

DF-FLOPS adds a weighting function to the existing FLOPS loss and introduces no new entities. The disclosed free parameters are activation cutoffs (alpha, beta), regularization strength (lambda), a pruning threshold, and a periodic DF-estimation schedule. The schedule is the most consequential: the paper itself shows the static variant fails, so the dynamic updating, not merely the weighting, is load-bearing. The axioms are standard background in IR plus two ad hoc choices (the activation function and the estimation schedule).

free parameters (6)
  • alpha (activation cutoff) = 0.1
    Chosen via pilot study on MS-Marco dev; sets the DF/|C| ratio above which terms are strongly penalized. The method's latency and effectiveness tradeoff depends on this value.
  • beta (activation steepness) = 10
    Chosen in the same pilot; controls how sharply the penalty rises for terms crossing alpha. The high value concentrates the penalty on the most frequent terms.
  • lambda peak (regularization strength) = 10^-1 to 10^3 for DF-FLOPS; 10^-3 to 10^1 for FLOPS
    The regularization factor is quadratically increased to a peak value tuned per model; the authors report diminishing returns for higher FLOPS lambda and use larger lambda for DF-FLOPS because its loss is bounded by FLOPS.
  • DF estimation interval = every 100 training steps
    DF weights are refreshed from validation vectors every 100 steps; the static precomputed DF variant performs considerably worse, so this schedule is load-bearing.
  • DF validation sample size = 50K query-passage pairs
    Size of the sample used for periodic DF estimation; not ablated in the paper.
  • inference pruning threshold k = 150
    Top-k pruning applied at inference (pruning@150) for the production efficiency comparison; this contributes to the 10x latency claim.
assumptions (5)
  • standard math FLOPS minimization gives each term equal likelihood when minimized (per Paria et al. [24]).
    Invoked as background in Section 2; motivates modifying FLOPS rather than replacing it. The authors observe the theoretical result fails in practice for LSR models.
  • domain assumption Lower document frequency per token translates directly to lower retrieval latency because posting-list traversal dominates scoring cost in production engines.
    Stated in the abstract and Section 1; the entire efficiency argument rests on this relationship, which is plausible but not formally demonstrated here.
  • ad hoc to paper Periodic DF estimates from a 50K validation sample every 100 training steps are a sufficient proxy for corpus-wide DF and yield stable training.
    Section 2 'Approximating DF'; chosen for computational feasibility. The static precomputed DF variant is much worse, so this schedule is load-bearing.
  • ad hoc to paper The generalized logistic activation (Eq. 3) with alpha=0.1 and beta=10 shapes penalties as intended.
    Section 2 'Activation and Tuning'; as typeset, the formula is arithmetically inconsistent with its description, so the actual implementation must differ from the text.
  • domain assumption SPLADE-Doc (binary query tokens, DistilBERT base document encoder, MS-Marco training) is a representative production-oriented LSR setup.
    Section 3 'Training Details' follows [8]; conclusions may not transfer to query-encoder SPLADE variants, which the authors list as future work.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An Alternative to FLOPS Regularization to Effectively Productionize SPLADE-Doc." pith.science (2026). https://pith.science/paper/QIRGPJBH

@misc{pith2026250515070,
  author       = {Pith},
  title        = {Pith review of: An Alternative to FLOPS Regularization to Effectively Productionize SPLADE-Doc},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QIRGPJBH}},
  note         = {Machine review of arXiv:2505.15070}
}
read the original abstract

Learned Sparse Retrieval (LSR) models encode text as weighted term vectors, which need to be sparse to leverage inverted index structures during retrieval. SPLADE, the most popular LSR model, uses FLOPS regularization to encourage vector sparsity during training. However, FLOPS regularization does not ensure sparsity among terms - only within a given query or document. Terms with very high Document Frequencies (DFs) substantially increase latency in production retrieval engines, such as Apache Solr, due to their lengthy posting lists. To address the issue of high DFs, we present a new variant of FLOPS regularization: DF-FLOPS. This new regularization technique penalizes the usage of high-DF terms, thereby shortening posting lists and reducing retrieval latency. Unlike other inference-time sparsification methods, such as stopword removal, DF-FLOPS regularization allows for the selective inclusion of high-frequency terms in cases where the terms are truly salient. We find that DF-FLOPS successfully reduces the prevalence of high-DF terms and lowers retrieval latency (around 10x faster) in a production-grade engine while maintaining effectiveness both in-domain (only a 2.2-point drop in MRR@10) and cross-domain (improved performance in 12 out of 13 tasks on which we tested). With retrieval latencies on par with BM25, this work provides an important step towards making LSR practical for deployment in production-grade search engines.

Figures

Figures reproduced from arXiv: 2505.15070 by the authors.

Figure 1
Figure 1. FLOPS (a) enforces sparsity within documents, but not over terms. DF-FLOPS (b) addresses this by enforcing sparsity over both documents and terms. address this issue with static pruning [12] and dynamic pruning with distillation [25]. However, these methods do not study the effects of DF on latency. Although optimizations such as BlockMaxWAND [6] allow for skipping portions of lengthy posting lists, full posting lis… view at source ↗
Figure 2
Figure 2. Top 100 DF% in representations produced by SPLADE-Doc with FLOPS and DF-FLOPS on a sample of 100K passages. reasonably up-to-date DFs while avoiding extra GPU computation. We recognize that other strategies could be used to produce DF es￾timates (e.g., running averages during training), but we leave these as possible explorations for future work. Activation and Tuning: We propose using the following varia￾tion of th… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 7 canonical work pages

  1. [1]

    Steven Bird. 2006. NLTK: The Natural Language Toolkit. In ACL 2006, 21st International Conference on Computational Linguistics and 44th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference, Sydney, Australia, 17-21 July 2006, Nicoletta Calzolari, Claire Cardie, and Pierre Isabelle (Eds.). The Association for Compu...

  2. [3]

    Sebastian Bruch, Franco Maria Nardini, Cosimo Rulli, and Rossano Venturini

  3. [5]

    Pairing Clustered Inverted Indexes with𝜅-NN Graphs for Fast Approximate Retrieval over Learned Sparse Representations. InProceedings of the 33rd ACM International Conference on Information and Knowledge Management, CIKM 2024, Boise, ID, USA, October 21-25, 2024, Edoardo Serra and Francesca Spezzano (Eds.). ACM, 3642–3646. doi:10.1145/3627673.3679977

  4. [6]

    Shuai Ding and Torsten Suel. 2011. Faster top-k document retrieval using block-max indexes. In Proceeding of the 34th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2011, Beijing, China, July 25-29, 2011, Wei-Ying Ma, Jian-Yun Nie, Ricardo Baeza-Yates, Tat-Seng Chua, and W. Bruce Croft (Eds.). ACM, 993–1002....

  5. [7]

    Nick Craswell, Bhaskar Mitra, Emine Yilmaz, Daniel Campos, and Ellen M Voorhees. 2020. Overview of the TREC 2019 deep learning track.arXiv preprint arXiv:2003.07820 (2020)

  6. [8]

    Thibault Formal, Carlos Lassance, Benjamin Piwowarski, and Stéphane Clinchant

  7. [9]

    William Falcon and The PyTorch Lightning team. 2019. PyTorch Lightning. doi:10.5281/zenodo.3828935

  8. [10]

    Carlos Lassance and Stéphane Clinchant. 2022. An Efficiency Study for SPLADE Models. In SIGIR ’22: The 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, Madrid, Spain, July 11 - 15, 2022 , Enrique Amigó, Pablo Castells, Julio Gonzalo, Ben Carterette, J. Shane Culpepper, and Gabriella Kazai (Eds.). ACM, 2220–2226...

Show all 31 references
  1. [11]

    Carlos Lassance, Hervé Déjean, Thibault Formal, and Stéphane Clinchant

  2. [13]

    Carlos Lassance, Simon Lupart, Hervé Déjean, Stéphane Clinchant, and Nicola Tonellotto. 2023. A Static Pruning Study on Sparse Neural Retrievers. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2023, Tai...

  3. [14]

    Jimmy Lin and Xueguang Ma. 2021. A Few Brief Notes on DeepImpact, COIL, and a Conceptual Framework for Information Retrieval Techniques. CoRR abs/2106.14807 (2021). arXiv:2106.14807 https://arxiv.org/abs/2106.14807

  4. [15]

    arXiv:2403.06789 [cs.IR] https://arxiv.org/abs/2403.06789

    SPLADE-v3: New baselines for SPLADE. arXiv:2403.06789 [cs.IR] https://arxiv.org/abs/2403.06789

  5. [16]

    Carlos Lassance, Simon Lupart, Hervé Déjean, Stéphane Clinchant, and Nicola Tonellotto. 2023. A static pruning study on sparse neural retrievers. InProceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval. 1771–1775

  6. [17]

    Joel Mackenzie, Andrew Trotman, and Jimmy Lin. 2021. Wacky Weights in Learned Sparse Representations and the Revenge of Score-at-a-Time Query Evaluation. CoRR abs/2110.11540 (2021). arXiv:2110.11540 https://arxiv.org/abs/2110.11540

  7. [18]

    Antonio Mallia, Omar Khattab, Torsten Suel, and Nicola Tonellotto. 2021. Learning Passage Impacts for Inverted Indexes. InSIGIR ’21: The 44th International ACM SIGIR Conference on Research and Development in Information Retrieval, Virtual Event, Canada, July 11-15, 2021 , Fern...

  8. [19]

    Sean MacAvaney and Craig Macdonald. 2022. A Python Interface to PISA!. In SIGIR ’22: The 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, Madrid, Spain, July 11 - 15, 2022 , Enrique Amigó, Pablo Castells, Julio Gonzalo, Ben Carteret...

  9. [20]

    Sean MacAvaney, Franco Maria Nardini, Raffaele Perego, Nicola Tonellotto, Nazli Goharian, and Ophir Frieder. 2020. Expansion via Prediction of Importance with Contextualization. In Proceedings of the 43rd International ACM SIGIR conference on research and development in Inform...

  10. [21]

    Thong Nguyen, Mariya Hendriksen, Andrew Yates, and Maarten de Rijke. 2024. Multimodal Learned Sparse Retrieval with Probabilistic Expansion Control. In Advances in Information Retrieval - 46th European Conference on Information Retrieval, ECIR 2024, Glasgow, UK, March 24-28, 2...

  11. [22]

    Thong Nguyen, Sean MacAvaney, and Andrew Yates. 2023. A Unified Framework for Learned Sparse Retrieval. InAdvances in Information Retrieval - 45th European Conference on Information Retrieval, ECIR 2023, Dublin, Ireland, April 2-6, 2023, Proceedings, Part III (Lecture Notes in...

  12. [23]

    Antonio Mallia, Michal Siedlaczek, Joel Mackenzie, and Torsten Suel. 2019. PISA: Performant Indexes and Search for Academia. InProceedings of the Open-Source IR Replicability Challenge co-located with 42nd International ACM SIGIR Conference on Research and Development in Infor...

  13. [24]

    Antonio Mallia, Torsten Suel, and Nicola Tonellotto. 2024. Faster Learned Sparse Retrieval with Block-Max Pruning. InProceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2024, Washington DC, USA, July 14-18, 202...

  14. [25]

    Yifan Qiao, Yingrui Yang, Shanxiu He, and Tao Yang. 2023. Representation sparsification with hybrid thresholding for fast splade-based document retrieval. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2329–2333

  15. [26]

    Francis J Richards. 1959. A flexible growth function for empirical use. Journal of experimental Botany 10, 2 (1959), 290–301

  16. [27]

    Tri Nguyen, Mir Rosenberg, Xia Song, Jianfeng Gao, Saurabh Tiwary, Rangan Majumder, and Li Deng. 2016. MS MARCO: A Human Generated MAchine Reading COmprehension Dataset. InProceedings of the Workshop on Cognitive Computation: Integrating neural and symbolic approaches 2016 co-...

  17. [28]

    Biswajit Paria, Chih-Kuan Yeh, Ian En-Hsu Yen, Ning Xu, Pradeep Ravikumar, and Barnabás Póczos. 2020. Minimizing FLOPs to Learn Efficient Sparse Representations. In 8th International Conference on Learning Representa- tions, ICLR 2020, Addis Ababa, Ethiopia, April 26-30, 2020 ...

  18. [29]

    Hansi Zeng, Julian Killingback, and Hamed Zamani. 2025. Scaling Sparse and Dense Retrieval in Decoder-Only LLMs. arXiv:2502.15526 [cs.IR] https://arxiv.org/abs/2502.15526

  19. [30]

    Shengyao Zhuang and Guido Zuccon. 2021. TILDE: Term Independent Likelihood moDEl for Passage Re-ranking. InSIGIR ’21: The 44th International ACM SIGIR Con- ference on Research and Development in Information Retrieval, Virtual Event, Canada, July 11-15, 2021, Fernando Diaz, Chi...

  20. [31]

    Nandan Thakur, Nils Reimers, Andreas Rücklé, Abhishek Srivastava, and Iryna Gurevych. 2021. BEIR: A Heterogenous Benchmark for Zero-shot Evaluation of Information Retrieval Models. CoRR abs/2104.08663 (2021). arXiv:2104.08663 https://arxiv.org/abs/2104.08663

  21. [32]

    Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Rémi Louf, Morgan Fun- towicz, and Jamie Brew. 2019. HuggingFace’s Transformers: State-of-the-art Natural Language Processing. CoRR abs/1910.03771 (2019). arXiv...

  22. [2021]

    CoRR abs/2109.10086 (2021)

    SPLADE v2: Sparse Lexical and Expansion Model for Information Retrieval. CoRR abs/2109.10086 (2021). arXiv:2109.10086 https://arxiv.org/abs/2109.10086

  23. [2024]

    Efficient Inverted Indexes for Approximate Retrieval over Learned Sparse Representations. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2024, Washington DC, USA, July 14-18, 2024, Grace Hui Yang, Hongn...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.