Pith. sign in

REVIEW 3 major objections 6 minor 35 references

Reproducing LightMem: Naive RAG Is Just as Good for Memory Management

T0 review · 3 major / 6 minor · reviewed 2026-08-03 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This reproduction study finds that retriever choice, not memory construction, is the main driver of answer accuracy in long-term conversational memory systems, and that raw-turn retrieval generally matches or beats constructed memories.

desk verdict A careful reproduction showing retriever choice swamps memory-construction gains; the oracle-based construction-loss claim is the least solid piece, but the matched-depth results stand. read the letter →

arxiv 2607.29104 v1 pith:5ZLJE5QY submitted 2026-07-31 cs.IR

Yongjie Zhou (1) , Shuai Wang (1) , Bevan Koopman (1 , 2) , Guido Zuccon (1 , 3) ((1) The University of Queensland , (2) CSIRO , (3) Google) This is my paper · ORCID
classification cs.IR
keywords LLMagentmemoryconstructionretrieval-augmentedgenerationshort-termandlong-termreproducibilitystudyretrieverchoiceoracleevaluationtokenbudget
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

This reproduction study asks whether transforming a conversation history into compact memory entries (LightMem) is actually better than simply retrieving raw user turns (Naive RAG) when an LLM agent needs to answer a question. The authors recover the reported configuration trend of LightMem, but find that the choice of retriever swings answer accuracy over a fixed memory store from 58.1% to 75.5%. When retrieval depth is matched, naive retrieval over raw turns generally outperforms retrieval over constructed memories; constructed memories help mainly when the answer-generation budget is very tight. An oracle comparison shows construction drops answer-relevant information (77.7% vs 89.0% oracle accuracy). The paper's claim is that memory construction is a conditional context-efficiency trade-off, not a general advantage, and that retrieval deserves as much attention as construction.

What carries the argument

The evaluation framework isolates construction from retrieval by (1) fixing a LightMem memory store and varying 11 sparse, dense, and hybrid retrievers, (2) comparing LightMem and Naive RAG under matched retrieval depths and matched answering-token budgets, and (3) using oracle conditions (providing gold evidence) to remove retrieval error. This separation lets the authors attribute accuracy differences to memory construction versus retrieval, rather than to end-to-end system behavior.

What would settle it

Re-annotate the benchmark's gold evidence at the level of constructed memory entries, or test a construction method whose oracle accuracy matches the raw-turn oracle (about 89%); if either yields oracle parity, the claim that construction removes answer-relevant information would be refuted as a general property.

Watch

Extended reading notes

Core claim

The central claim is that memory construction does not provide a consistent advantage over retrieving the original dialogue turns. At matched retrieval depths, Naive RAG generally performs better; at matched answering-token budgets, LightMem performs better only when the budget is tight (about 330 tokens) and the advantage disappears by about 935 tokens. Oracle evaluation further shows that memory construction removes some answer-relevant information: when given gold evidence, Naive RAG reaches 89.0% accuracy versus 77.7% for LightMem. The best real retriever over the constructed store reaches 75.5%, approaching but not reaching the constructed-memory oracle. The authors conclude that LightM

Load-bearing premise

The oracle and recall comparisons assume that each gold raw turn maps to exactly the memory entries constructed or updated from that turn via the has_answer flag; if this mapping is systematically lossy or noisy, the measured 'information lost during construction' could be an artifact of the mapping rather than genuine information loss.

Editorial extensions

If this is right

  • Retriever sensitivity: changing only the retriever over the same LightMem store shifts answer accuracy from 58.1% to 75.5%, so any memory-construction claim should be evaluated across retrievers.
  • At matched retrieval depths (top-3/5/10), Naive RAG outperforms LightMem for most retrievers; constructed memories are not a universally better representation.
  • LightMem's advantage is concentrated under tight answering-token budgets (about 330 tokens per question) and fades as budgets grow to about 935 tokens.
  • Oracle evaluation shows construction discards answer-relevant information (77.7% vs 89.0% oracle accuracy), and the best real retriever (75.5%) approaches but does not reach the constructed-memory oracle.
  • LightMem's construction cost (about 120k tokens and 117 LLM calls per sample) is only recouped in answering-token savings after roughly 321 questions, so the trade-off depends on conversation length.

Reading between the lines

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

  • A practical corollary the authors leave implicit: for agents whose conversations are short or where users rarely return, the amortization math favours skipping construction entirely and investing in a strong raw-turn retriever.
  • The token-budget finding suggests memory construction may be most valuable for deployment constraints (small context windows, per-token cost), not for accuracy per se; a testable extension is measuring LightMem-vs-Naive-RAG under a hard context-window limit with a long conversation.
  • The paper's stage-separation methodology could be applied to other memory systems (summarisation, knowledge graphs) to diagnose whether reported gains come from representation or retrieval.
  • An untested but natural next step is query formulation or rewriting on the raw-turn side, or reranking on the constructed side; the paper's data imply such retrieval-side interventions might close the residual gap to the raw-turn oracle.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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. This paper reproduces LightMem on LongMemEval-S (444 questions) and compares it with Naive RAG over raw user turns. RQ1 finds that LightMem's configuration ordering is reproduced but absolute accuracy and construction-cost values differ from the original report. RQ2 holds a fixed LightMem store (r=0.8, th=1024) and varies 11 retrievers, reporting large performance variation (58.1%–75.5% answer accuracy; Recall@10 0.390–0.587). RQ3 compares LightMem with Naive RAG at matched retrieval depths and matched answering-token budgets, finding that Naive RAG generally wins at matched depth, LightMem wins mainly at tight token budgets, and an oracle gap (89.0% vs 77.7%) is interpreted as evidence that memory construction removes answer-relevant information. The paper concludes that LightMem offers a conditional context-efficiency trade-off rather than a general advantage over Naive RAG.

Significance. If the main conclusions hold, the paper makes a valuable contribution: it challenges the common assumption that constructed memory representations are inherently superior to raw-turn retrieval, and it provides a controlled protocol for separating construction loss from retrieval error. Strengths include the use of the released LightMem implementation, evaluation with two different LLM judges that yield consistent conclusions, significance tests with multiple-comparison correction, and detailed appendix tables. The matched-depth and matched-budget comparisons (Figure 4, Tables 7–8) are informative and do not depend on the fragile oracle mapping. However, the mechanistic claim that construction removes answer-relevant information is built on an oracle condition whose provenance mapping is admitted to be incomplete, so the headline oracle gap needs additional validation before that explanation is fully established.

major comments (3)
  1. [§5.1, §6.2, Table 6] The LightMem oracle treats an entry as relevant only if it was 'constructed or updated from a turn marked has_answer'. The paper itself notes this is not a strict upper bound, and Table 4 shows real retrievers exceeding the oracle on single-session-user questions (e.g., Qwen3-Emb-0.6B: 93.8 vs Oracle 87.5). Consequently, the 11.3-point gap between the Naive RAG oracle (89.0%) and the LightMem oracle (77.7%) may partly reflect an incomplete provenance mapping rather than information lost during construction. Because the abstract and §6.2 use this gap to support the claim that construction 'removes some answer-relevant information', please (a) report the fraction of gold turns that map to at least one constructed memory entry, (b) manually audit a sample of constructed entries for answer-bearing content not linked to has_answer turns, and/or (c) recompute the oracle using all memory entrie
  2. [§5.2, Table 4] Several category-level claims, such as 'retriever choice matters most for multi-session and single-session-preference questions', are based on small subsets whose sizes are not reported in Table 4. For example, the Single-pref column shows BM25 at 33.3% and Qwen3-Emb-4B at 80.0%, a 46.7-point spread; with a category of a few dozen questions this is compatible with wide confidence intervals even when a significance test is nominally passed. Reporting N per category and confidence intervals (or standard errors) for the key accuracy cells would make the retriever-sensitivity claims more robust. This does not affect the overall ranking, but it does affect the paper's characterization of where retriever choice matters.
  3. [§6.2, Table 8] The matched-budget comparison shows a mean LightMem advantage of +5.5 at ~330 tokens, +2.2 at ~500, and -0.9 at ~935, but the distribution across retrievers is highly heterogeneous. At ~330t, cell-level differences range from -2.3 (Fusion Qwen3-0.6B+BM25) to +17.8 (all-MiniLM-L6-v2); at ~935t, several cells favour Naive RAG. The conclusion that LightMem is 'most useful under tight budgets' is fair as a central tendency, but the paper should also state the proportion of retrievers with positive deltas and avoid implying a uniform advantage, especially since §7 correctly concludes that the value depends on the retriever.
minor comments (6)
  1. [§3.1, Eq. (1)] N_comp ≈ rN is only approximate; the actual compressed length depends on LLMLingua-2's tokenization and the rate setting. Please clarify that r is a target compression parameter, not an exact post-compression token count.
  2. [Table 1] The abbreviation 'OP-update' is used without definition. Spell out 'offline update' or explain the abbreviation in the table caption or text.
  3. [Table 6] The 'Answering tokens / question' row reads '∼903∼529' with no separator. It should be '∼903 / ∼529'.
  4. [Figure 3] The oracle marker lies outside the displayed x-axis range (Recall@10=1.0). Consider adding an arrow or a break in the axis so the comparison with real retrievers is visually clear.
  5. [§4.2] The block-quote formatting for 'Original finding' and 'Reproduction target' is unusual and interrupts the narrative. Use normal prose for these statements.
  6. [Appendix B] The consistent ranking under the gpt-4o-mini judge is a strong robustness result. Add one sentence in the main text pointing to this verification, rather than leaving it only in the appendix.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: all claims are measured comparisons on an external benchmark; the two author-overlapping citations are non-load-bearing retriever baselines.

full rationale

This is a reproduction/measurement study, not a derivation chain. RQ1 compares against the original LightMem numbers; RQ2 varies retrievers over a fixed constructed store; RQ3 compares constructions against raw turns under matched depths and token budgets and under oracle conditions. No parameter is fitted to a subset of data and then reported as a prediction, and no result is equivalent by construction to its input. The oracle comparison (89.0% vs 77.7%, Table 6) is an empirical gold-evidence ceiling, and the paper explicitly discloses its limitation: 'It is not a strict upper bound, however, because entries not linked to a labelled turn may still contain useful context' (§5.1). The has_answer-to-memory provenance mapping is a benchmark-labeling caveat, not a circular reduction; the paper even reports retrievers exceeding the oracle on some categories (Table 4, single-session-user), which confirms the mapping's incompleteness rather than concealing it. The only self-citations are PromptReps [32] and score fusion [24], used as off-the-shelf retriever baselines; the main retriever-variation and naive-RAG-competitiveness findings also hold across many non-self-cited retrievers (BM25, SPLADE-v3, all-MiniLM-L6-v2, Qwen3-Embedding), so these citations are not load-bearing. A skeptic could question whether the oracle gap is inflated by a lossy provenance mapping, but that is a validity concern about the benchmark linkage, not circularity in the paper's own reasoning. Accordingly, no circular step is identified; score 2 reflects only the presence of two minor, non-load-bearing self-citations, not any circular argument.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The paper contributes an evaluation, not a derivation; it inherits LightMem's compression rate r and capacity threshold th from the original system and selects one configuration on the test set. The main load-bearing assumptions are the LLM judge's validity, the has_answer-to-memory relevance mapping for oracle/recall, and single-run determinism.

free parameters (1)
  • LightMem configuration (r, th) fixed for RQ2/RQ3 = (0.8, 1024)
    Selected as best-performing configuration on the same 444-question test set in RQ1 (§4.2), then used to build the fixed store for §5 and §6. Selecting on the evaluation set is a mild form of test-set peeking for the method under comparison; Naive RAG receives no equivalent tuning.
assumptions (4)
  • domain assumption LLM judge (gpt-5.5-2026-04-23) verdicts are a valid measure of answer correctness; gpt-4o-mini check confirms ranking only.
    Answer accuracy is the primary metric; no human-annotated gold correctness labels are used. §3.3, Table 2.
  • domain assumption A constructed memory entry is relevant iff it was constructed or updated from a raw turn labeled has_answer.
    Used for Recall@10 and oracle conditions in §5.1 and §6.1; the paper acknowledges this mapping is approximate and 'not a strict upper bound.'
  • domain assumption LLM generation (Qwen3-30B-A3B-Instruct-2507, vLLM) is sufficiently stable that single-run accuracy without confidence intervals supports the comparisons.
    No repeated generations or variance reporting in Tables 3-8; the paper relies on per-question paired tests across a single run.
  • domain assumption LongMemEval-S (444 graded questions after excluding single-session-assistant) is representative of long-term conversational memory evaluation.
    All conclusions are drawn from this benchmark; the exclusion follows the original study but restricts scope.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Reproducing LightMem: Naive RAG Is Just as Good for Memory Management." pith.science (2026). https://pith.science/paper/5ZLJE5QY

@misc{pith2026260729104,
  author       = {Pith},
  title        = {Pith review of: Reproducing LightMem: Naive RAG Is Just as Good for Memory Management},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5ZLJE5QY}},
  note         = {Machine review of arXiv:2607.29104}
}
read the original abstract

Long-term conversational agents require access to information from earlier interactions, such as a user's preferences, past requests, or previously mentioned facts. Repeatedly providing the full dialogue history can be expensive as conversations grow, so many memory approaches instead transform past interactions into compact entries that can be retrieved when needed. LightMem is a recent lightweight memory-management approach that reports strong effectiveness while maintaining relatively low construction cost. However, it still relies on a separate constructed memory representation and is evaluated with only one retriever, leaving unclear how sensitive its results are to retriever choice and whether memory construction discards answer-relevant information. In this study, we reproduce LightMem and compare it with Naive RAG, which retrieves directly from raw user turns. We recover LightMem's main configuration trend, but find that retriever choice is a major source of performance variation: changing only the retriever over a fixed LightMem store shifts answer accuracy from 58.1% to 75.5%. Constructed memories also do not consistently outperform raw-turn retrieval. Naive RAG generally performs better at matched retrieval depths, whereas LightMem performs better mainly under tight answering-token budgets. Oracle evaluation further shows that memory construction removes some answer-relevant information. Overall, LightMem offers a context-efficiency trade-off rather than a general advantage over Naive RAG. Its value depends on the retriever and available token budget, motivating future work on retrieval, reranking, query formulation, and their interaction with raw and constructed memory representations.

Figures

Figures reproduced from arXiv: 2607.29104 by the authors.

Figure 1
Figure 1. The original LightMem pipeline. Dialogue turns are [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of our evaluation framework. LightMem retrieves from constructed memory entries, whereas Naive RAG [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Relationship between Recall@10 and answer accuracy for LightMem at top- [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Comparison between LightMem and Naive RAG. Each cell reports the answer-accuracy difference, [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: GPT-4o-mini judge. Relationship between Recall@10 and answer accuracy for LightMem at top- [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: GPT-4o-mini judge. Controlled comparison between LightMem and Naive RAG. Each cell reports the answer-accuracy [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 13 linked inside Pith

  1. [1]

    Ding Chen, Simin Niu, Kehang Li, Peng Liu, Xiangping Zheng, Bo Tang, Xinchi Li, Feiyu Xiong, and Zhiyu Li. 2025. Halumem: Evaluating hallucinations in memory systems of agents.arXiv preprint arXiv:2511.03506(2025)

    arXiv 2025

  2. [2]

    Prateek Chhikara, Dev Khant, Saket Aryan, Taranjeet Singh, and Deshraj Ya- dav. 2025. Mem0: Building production-ready ai agents with scalable long-term memory.arXiv preprint arXiv:2504.19413(2025). 9 Yongjie Zhou, Shuai Wang, Bevan Koopman, and Guido Zuccon

    arXiv 2025

  3. [3]

    Gordon V Cormack, Charles LA Clarke, and Stefan Buettcher. 2009. Reciprocal rank fusion outperforms condorcet and individual rank learning methods. In Proceedings of the 32nd international ACM SIGIR conference on Research and development in information retrieval. 758–759

    2009

  4. [4]

    Jizhan Fang, Xinle Deng, Haoming Xu, Ziyan Jiang, Yuqi Tang, Ziwen Xu, Shumin Deng, Yunzhi Yao, Mengru Wang, Shuofei Qiao, Huajun Chen, and Ningyu Zhang

  5. [5]

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

  6. [6]

    Thibault Formal, Benjamin Piwowarski, and Stéphane Clinchant. 2021. Splade: Sparse lexical and expansion model for first stage ranking. InProceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2288–2292

    2021

  7. [7]

    Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Ji- awei Sun, Meng Wang, and Haofen Wang. 2023. Retrieval-augmented generation for large language models: A survey.arXiv preprint arXiv:2312.10997(2023)

    arXiv 2023

  8. [8]

    Yuanzhe Hu, Yu Wang, and Julian McAuley. 2025. Evaluating memory in llm agents via incremental multi-turn interactions.arXiv preprint arXiv:2507.05257 (2025)

    arXiv 2025

Show all 35 references
  1. [9]

    Jiazheng Kang, Mingming Ji, Zhe Zhao, and Ting Bai. 2025. Memory os of ai agent. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 25961–25970

    2025

  2. [10]

    Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. 2020. Dense passage retrieval for open- domain question answering. InProceedings of the 2020 conference on empirical methods in natural language processing (EMNLP)...

    2020

  3. [11]

    Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph Gonzalez, Hao Zhang, and Ion Stoica. 2023. Efficient memory management for large language model serving with pagedattention. In Proceedings of the 29th symposium on operating systems princip...

    2023

  4. [12]

    Carlos Lassance, Hervé Déjean, Thibault Formal, and Stéphane Clinchant. 2024. SPLADE-v3: New baselines for SPLADE.arXiv preprint arXiv:2403.06789(2024)

    2024 arXiv

  5. [13]

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rock- täschel, et al. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks.Advances in neural information processing ...

    2020

  6. [14]

    Zhiyu Li, Shichao Song, Hanyu Wang, Simin Niu, Ding Chen, Jiawei Yang, Chenyang Xi, Huayi Lai, Jihao Zhao, Yezhaohui Wang, et al. 2025. Memos: An operating system for memory-augmented generation (mag) in large language models.arXiv preprint arXiv:2505.22101(2025)

    2025 arXiv

  7. [15]

    Charles Packer, Vivian Fang, Shishir_G Patil, Kevin Lin, Sarah Wooders, and Joseph_E Gonzalez. 2023. MemGPT: towards LLMs as operating systems.arXiv preprint arXiv:2310.08560(2023). https://arxiv.org/abs/2310.08560

    2023 arXiv

  8. [16]

    Zhuoshi Pan, Qianhui Wu, Huiqiang Jiang, Xufang Luo, Hao Cheng, Dongsheng Li, Yuqing Yang, Chin-Yew Lin, H Vicky Zhao, Lili Qiu, et al. 2025. On mem- ory construction and retrieval for personalized conversational agents. InThe Thirteenth International Conference on Learning Re...

    2025

  9. [17]

    Zhuoshi Pan, Qianhui Wu, Huiqiang Jiang, Menglin Xia, Xufang Luo, Jue Zhang, Qingwei Lin, Victor Rühle, Yuqing Yang, Chin-Yew Lin, et al. 2024. Llmlingua-2: Data distillation for efficient and faithful task-agnostic prompt compression. In Findings of the Association for Comput...

    2024

  10. [18]

    Preston Rasmussen, Pavlo Paliychuk, Travis Beauvais, Jack Ryan, and Daniel Chalef. 2025. Zep: a temporal knowledge graph architecture for agent memory. arXiv preprint arXiv:2501.13956(2025)

    2025 arXiv

  11. [19]

    Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. InProceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCN...

    2019

  12. [20]

    Alireza Rezazadeh, Zichao Li, Wei Wei, and Yujia Bao. 2025. From isolated conversations to hierarchical schemas: Dynamic tree memory representation for llms. InThe Thirteenth International Conference on Learning Representations, Vol. 2025

    2025

  13. [21]

    2009.The probabilistic relevance frame- work: BM25 and beyond

    Stephen Robertson and Hugo Zaragoza. 2009.The probabilistic relevance frame- work: BM25 and beyond. Vol. 4. Now Publishers Inc

    2009

  14. [22]

    Sentence Transformers. [n. d.]. all-MiniLM-L6-v2. Hugging Face model reposi- tory. Accessed July 14, 2026. https://huggingface.co/sentence-transformers/all- MiniLM-L6-v2

    2026

  15. [23]

    Zhen Tan, Jun Yan, I-Hung Hsu, Rujun Han, Zifeng Wang, Long Le, Yiwen Song, Yanfei Chen, Hamid Palangi, George Lee, et al. 2025. In prospect and retrospect: Reflective memory management for long-term personalized dialogue agents. InProceedings of the 63rd Annual Meeting of the...

    2025

  16. [24]

    Shuai Wang, Shengyao Zhuang, and Guido Zuccon. 2021. Bert-based dense retrievers require interpolation with bm25 for effective passage retrieval. InPro- ceedings of the 2021 ACM SIGIR international conference on theory of information retrieval. 317–324

    2021

  17. [25]

    Yuxin Wang, Paul Thomas, Zhiwei Yu, Yuan Gao, Saeed Hassanpour, Soroush Vosoughi, Robert Sim, and Nick Craswell. 2026. Memory Makes the Difference: Evaluating How Different Memory Roles Shape Conversational Agents.arXiv preprint arXiv:2606.25361(2026)

    2026 arXiv

  18. [26]

    Di Wu, Hongwei Wang, Wenhao Yu, Yuwei Zhang, Kai-Wei Chang, and Dong Yu. 2025. Longmemeval: Benchmarking chat assistants on long-term interactive memory. InThe Thirteenth International Conference on Learning Representations. https://openreview.net/forum?id=pZiyCaVuti

    2025

  19. [27]

    Jianfei Xiao, Xiang Yu, Chengbing Wang, Wuqiang Zheng, Xinyu Lin, Kaining Liu, Hongxun Ding, Yang Zhang, Wenjie Wang, Fuli Feng, et al. 2026. AlpsBench: An LLM Personalization Benchmark for Real-Dialogue Memorization and Preference Alignment.arXiv preprint arXiv:2603.26680(2026)

    2026 arXiv

  20. [28]

    Wujiang Xu, Zujie Liang, Kai Mei, Hang Gao, Juntao Tan, and Yongfeng Zhang

  21. [29]

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. 2025. Qwen3 technical report.arXiv preprint arXiv:2505.09388(2025)

    2025 arXiv

  22. [30]

    Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, et al. 2025. Qwen3 embedding: Advancing text embedding and reranking through foundation models.arXiv preprint arXiv:2506.05176(2025)

    2025 arXiv

  23. [31]

    Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. 2024. Memo- rybank: Enhancing large language models with long-term memory. InProceedings of the AAAI conference on artificial intelligence, Vol. 38. 19724–19731

    2024

  24. [32]

    Shengyao Zhuang, Xueguang Ma, Bevan Koopman, Jimmy Lin, and Guido Zuc- con. 2024. PromptReps: Prompting large language models to generate dense and sparse representations for zero-shot document retrieval. InProceedings of the 2024 Conference on Empirical Methods in Natural Lan...

    2024

  25. [2021]

    arXiv preprint arXiv:2109.10086(2021)

    SPLADE v2: Sparse lexical and expansion model for information retrieval. arXiv preprint arXiv:2109.10086(2021)

    2021 arXiv

  26. [2025]

    A-mem: Agentic memory for llm agents.Advances in Neural Information Processing Systems38 (2025), 17577–17604

    2025

  27. [2026]

    In The Fourteenth International Conference on Learning Representations

    Lightmem: Lightweight and efficient memory-augmented generation. In The Fourteenth International Conference on Learning Representations. https: //arxiv.org/abs/2510.18866

    arXiv

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.