Pith. sign in

REVIEW 4 major objections 4 minor 47 references

KV-Skill: Forging Expertise in the Model's Native Language

T0 review · 4 major / 4 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read The paper claims that task knowledge can be stored outside both the prompt and the model's weights, as an external factorized operator that a frozen language model reads through a small trained interface.

desk verdict A well-crafted empirical paper on external task operators with a strong matched reward study, but the registration path needs a no-content control before the headline claim fully lands. read the letter →

arxiv 2608.05475 v1 pith:3R3YEAI5 submitted 2026-08-05 cs.LG

Zhaowei Han , Xiang Zhang , Bing Han , Kai Liu , Danqi Hu , Jie Liu This is my paper · ORCID
classification cs.LG
keywords KV-Skillexternalfactorizedoperatorfrozenlanguagemodeltaskknowledgetext-derivedskillrewardlearningresidualstreamsteeringrank-onecompression
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

KV-Skill is a design space for storing task knowledge outside both the prompt and the backbone, in an external factorized operator that a frozen language model reads through a lightweight interface. The paper tries to establish that the same procedural knowledge becomes more effective when converted from text into this operator form, and that an unwritten procedure can be learned from reward directly into a compact latent operator. Across ten benchmarks and four backbones, converting a text skill to a KV-Skill consistently improves over the text skill, and under matched reward and parameter budgets the operator beats soft prefixes, prefix KV, and LoRA in seven of eight settings. A registered text-derived operator can be compressed to one task-aligned direction per injection layer with little loss, and one shared interface can retain several independently loadable skills. The conclusion is that task knowledge can be acquired from text or experience, compressed, and deployed separately from the model's weights.

What carries the argument

The central object is an external factorized associative operator $M_s^{(\ell)} = W_s^{(\ell)} U_s^{(\ell)\,\top}$ read at selected layers $\ell$: the columns of $U$ act as latent keys that measure the signed relevance of each skill slot to the current residual query, and the columns of $W$ return the corresponding response. The query is built by an interface $I$ from the model's residual state, and the operator's response is injected back through a gated residual branch, adding no prompt positions and no attention-KV-cache entries. Registration and reward learning are two endpoints of one design space: registration compiles an authored text skill into a fixed, token-aligned operator from a single prefill, while reward learning develops a compact latent operator shared across depths from task outcomes.

What would settle it

A decisive test is to register a skill whose single-prefill residual states do not cover the held-out query distribution, and compare the rank-one operator against a magnitude-matched random direction: the central claim predicts the task-aligned direction retains most of the gain while the random direction collapses, so observing the opposite pattern would refute the representativeness assumption.

Watch

Extended reading notes

Core claim

The paper's central claim is that task knowledge can live outside both the prompt and the weights, as an external factorized associative operator $M_s = W_s U_s^\top$ that is read from the residual stream through a per-backbone interface $I$. Registration sets the operator's factors to the normalized residual states harvested from a single frozen prefill of an authored text skill, and then trains only the interface so the frozen model can use the fixed operator. Reward learning instead optimizes a compact 32-slot latent operator directly from verifier outcomes, with or without a text-informed initialization. The evidence includes a LiveMath registration accuracy of 77.2 versus 23.4 for the text skill and 64.5 for SoftSkill, best results in seven of eight matched reward-training comparisons against soft prefixes, prefix KV, and LoRA, and a rank analysis showing that the operator retains 90 to 100 percent of its gain at one task-aligned direction per layer while matched random directions fail. This is taken to show that the representation through which task knowledge is delivered, not just the objective, determines how well a frozen model can execute a procedure.

Load-bearing premise

Registration assumes that a single frozen pass through the authored text skill produces residual states that already cover the task's query distribution; if those states are unrepresentative, training the interface cannot recover the lost procedure.

Editorial extensions

If this is right

  • Because the operator is read from a separate residual branch, a KV-Skill consumes no prompt positions and does not grow with conversation length, so the context window is preserved.
  • A KV-Skill can be loaded, swapped, or removed without modifying backbone weights, making each skill an independently deployable capability object.
  • Reward learning can discover an effective operator from verifier feedback alone, so tasks without an authored procedure can still gain an external skill.
  • The rank-one compression result implies that a text-derived skill can be stored as roughly one direction per layer after a one-time SVD, provided the interface stays fixed.
  • Because the shared per-backbone interface can retain several registered skills without measurable forgetting, a single reader can serve a growing skill library.

Reading between the lines

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

  • The registration result suggests a compiler view of text-to-operator conversion: the paper's evidence implies that the retained object is a query-dependent steering direction, and one testable extension is to build rank-one KV-Skills directly from text without first materializing the full token-aligned operator.
  • If the operator generalizes across query distributions, this external representation could be used for continual learning without weight updates; the paper only demonstrates retention across three tasks on one backbone, so cross-task scaling remains an open extension.
  • The sparse-terminal-reward result on ALFWorld implies that the bottleneck is credit assignment rather than operator capacity, so a dense-reward version of reward learning is a natural next test for long-horizon tasks.
  • The paper does not test transferring one KV-Skill tensor across backbones; a direct extension would ask whether a compact reward-learned operator learned in one model's skill space transfers to another model after only a new interface.
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

4 major / 4 minor

Summary. The paper introduces KV-Skill, an external factorized associative operator Ms (with factors U_s, W_s) that a frozen language model reads through a lightweight trainable interface I. The authors propose two construction paths: registration, which converts an authored text skill into a fixed operator Ms = H_s H_s^T using residual states from a single prefill and then trains only the interface with a combination of KL matching to the text teacher and cross-entropy on gold answers; and reward learning, which jointly optimizes a compact operator and interface directly from verifier outcomes. Experiments across ten benchmarks and four backbones report that registration substantially improves over the source text skill, that reward-learned KV-Skills outperform soft prefixes, prefix-KV, and LoRA under matched reward training and parameter budgets, and that text-derived operators can be truncated to one task-aligned direction per layer with little loss. The paper also demonstrates that one shared interface can load multiple skills and that swapping the loaded operator changes task behavior.

Significance. If the central claims hold, the paper offers a genuinely useful design space for storing and loading task knowledge outside both the prompt and the model weights, with modularity benefits and a clear separation between knowledge acquisition and deployment. The matched reward comparisons in Table 5 are well controlled and provide strong evidence for a substrate effect; the swap experiment in Table 4 gives direct evidence that, after training, the loaded operator, not the interface, carries task identity. The rank-one compression result is surprising and mechanistically informative. The paper also ships a thorough resource-accounting appendix and is generally careful with the limitations of its claims. However, the registration path—which anchors the abstract's central claim that 'task knowledge can be acquired from text ... compressed into an external operator'—lacks a no-content control that would separate the contribution of the text-derived operator's content from the addition of a trained interface with gold-label supervision.

major comments (4)
  1. [§4.1 and §3.4 (Eq. 10)] The registration comparison is confounded: the KV-Skill condition trains an interface with gold labels plus KL to the text teacher, while the Text Skill and other baselines receive no equivalent training on a fixed operator. To support the claim that converting text to a KV-Skill makes the same procedural knowledge more effective, the authors must run a no-content control: train the same interface with the same registration loss (Eq. 10) on a random fixed operator of the same shape (or an operator derived from an unrelated text skill) and report the resulting performance. Table 4 and Figure 4 are post-hoc controls: they hold an interface that was trained on the correct operator fixed and then swap or truncate the operator. They do not test whether the text-derived content is necessary to achieve the gain when the interface is trained from scratch. If a random operator under the same training protocol reaches comparable accuracy, the registration headline collapses to 'training a small interface with gold labels helps' and the text-derived mechanism is not supported.
  2. [§3.3 (Eq. 8) and §3.4] The text-derived operator M_s^ℓ = H̄_s^ℓ H̄_s^ℓ⊤ assumes that a single frozen prefill of the authored skill produces residual states that form a sufficient fixed basis for the task. This representativeness assumption is load-bearing for the registration path: if the queries encountered during interface training or test lie outside the span of the collected text-skill states, the operator cannot express the required response regardless of how much the interface is trained. The paper does not test this assumption directly. The random-operator control proposed above would also serve as a partial test of whether the text states' specific content matters, but the authors should additionally consider ablating the number of prefills or the length of the skill document to probe whether the operator improves with more representative states.
  3. [§B.1 and §4.2 (c_k = (k/N_s)^(1/4))] The rank-correction exponent c_k = (k/N_s)^(1/4) is described as 'analytic' and 'parameter-free', but no derivation is provided. If the fourth-root law was chosen post hoc to make the rank-one results align with the full-rank operator, the claim of no fitted coefficient is misleading. The authors should provide a derivation from the read scale (e.g., from the expected norm of U^⊤ q when U contains N_s normalized states) or, failing that, report the sensitivity of the rank-one accuracy to the exponent (e.g., 1/2, 1/3, 1/5). This matters because the rank-one compression result is one of the paper's headline findings.
  4. [§4.5 and Table 10] The sequential-registration result, used to support the claim that one shared interface can retain multiple independently loadable KV-Skills without measurable forgetting, is based on a single seed, and the LiveMath intermediate value is reported as a range rather than a point estimate. This is an inherently high-variance claim (small task, one seed, sequential optimization). The replay ablation is also a single-run result. The authors should provide at least three seeds for the sequential registration experiment or explicitly soften the 'no measurable forgetting' claim to a preliminary observation.
minor comments (4)
  1. [Table 5] The parameter counts are not exactly matched: KV-Skill uses 1.330M and 1.067M parameters on the two backbones, while LoRA uses 1.376M and 1.024M respectively. The text says 'matched parameter count' but the numbers differ by up to 4%; please clarify the matching tolerance or state that the match is approximate.
  2. [Table 3 and §B.1] The 'Random direction' row in Table 3 reports LiveMath accuracy 26.6, which is below the Base accuracy of 22.4 in Table 1; the paper later says 'a magnitude-matched random direction preserves at most 22% and returns SearchQA to Base' but the LiveMath number is not Base. Please add a footnote to Table 3 clarifying the Base value under the same generation budget and example count.
  3. [Abstract and §4.1] The abstract says 'across ten benchmarks', but the main tables report results on five main tasks, and the remaining benchmarks (STaRK-Prime, STaRK-MAG, OfficeQA, SpreadsheetBench, ALFWorld) are only in the appendix. Consider adding a sentence in the main text pointing to the appendix for the full ten-benchmark list.
  4. [§C.5 and §C.6] The resource-accounting section is thorough, but the sentence 'The cached-swap measurement covers changing the operator used for future tokens. It does not recompute representations already stored in the model’s attention KV cache' should be moved earlier in the main text (near §4.3) so that readers do not over-interpret the '0.00 ms' swap time as including history re-encoding.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central KV-Skill claims are empirical, externally benchmarked, and do not reduce by construction to their inputs.

full rationale

The paper's load-bearing assertions are empirical: registration and reward learning are measured against Text Skill, SkillOpt, SoftSkill, soft prefixes, prefix KV, and LoRA under matched budgets. The text-derived operator is defined by Eqs. 8–9 as the self-associative outer product of residual states from a single prefill, and the interface is trained by Eq. 10 with KL and gold CE; neither equation makes the reported accuracy gains true by definition. The gains are contingent measurements, and the SkillOpt/SoftSkill comparisons provide external baselines that also receive task training, so the headline comparison is not a fitted parameter renamed as a prediction. The rank and direction analysis is post-hoc but controlled: the top SVD direction is selected from the operator, not from task accuracy, and the matched-random-direction control fails, giving independent evidence of specificity; the stated analytic correction c_k = (k/N_s)^{1/4} is presented as coefficient-free, not fitted to the reported scores. There is no load-bearing self-citation, no uniqueness theorem imported from the authors' prior work, and no ansatz smuggled in via citation; the paper explicitly disclaims novelty of the associative operator form. The absence of a no-content random-operator registration control is a potential experimental confound about what drives the registration gain, but that is a validity concern rather than circularity: the paper does not derive the gain from the operator definition or from a self-citation chain. Accordingly, no step in the derivation chain is equivalent to its own input.

Assumptions & free parameters 3 free parameters · 5 assumptions · 1 invented entities

Axiom ledger summary: the paper introduces no unobservable entities; the operator and interface are concrete, testable tensors. The main postulates are that residual-stream injection is sufficient for task steering, that one text prefill yields a sufficient operator basis, and that a fourth-root scale correction is correct. The compact dimensions and layer set are hand-chosen but swept or fixed. The central performance claims do not reduce to fitted quantities.

free parameters (3)
  • Rank correction exponent c_k = (k/N_s)^(1/4) = exponent 1/4, no fitted coefficient claimed
    Introduced in Section 4.2 to keep read scale comparable across SVD ranks in the rank-one compression analysis. Called analytic, but the derivation is not shown, so it functions as a hand-chosen correction for that result.
  • Compact operator dimensions (m_s, d_s) = m_s=32, d_s=256
    Chosen design hyperparameters for the reward-learned operator. Appendix B.3 sweeps both and shows stability, so the central claim does not depend on the exact values, but they are still hand-chosen.
  • Injection layer set D = 4 layers, exact layer indices not specified in main text
    The number and choice of injection depths is a design decision with no ablation. The central results are produced with this fixed set.
assumptions (5)
  • domain assumption Task behavior can be steered by an additive residual update at selected layers (Eq. 6).
    The whole interface read is an additive injection h'_l = h_l + ...; the framework assumes this is sufficient to change task behavior without modifying attention or weights.
  • ad hoc to paper One frozen prefill of the text skill yields residual states that are a sufficient basis for the operator (Eq. 8).
    U=W=H fixes the operator from a single pass over the skill document; this representativeness assumption is not tested directly.
  • domain assumption The text-conditioned model is a good teacher for the operator-conditioned model (Eq. 10).
    Registration uses KL(p_text || p_KV); standard distillation assumption, not independently verified in this setting.
  • standard math Group-relative reward normalization gives a valid learning signal (Eq. 13).
    Baseline subtraction within a group is a standard policy-gradient technique; assumes rewards are comparable across responses to the same input.
  • ad hoc to paper The rank-k read scale correction c_k follows a fourth-root law.
    The analytic correction in Section 4.2 is asserted without derivation; the rank-one compression result depends on it.
invented entities (1)
  • KV-Skill operator M_s and interface I independent evidence
    purpose: Stores procedural task knowledge outside the prompt and backbone and reads it through a trained residual branch.
    The operator is a concrete tensor object, not an unobservable. The swap experiment in Table 4 and the rank and direction controls provide falsifiable handles, and the code release allows external reproduction.

how reviews work

0 comments
Cite this review

Pith. "Pith review of KV-Skill: Forging Expertise in the Model's Native Language." pith.science (2026). https://pith.science/paper/3R3YEAI5

@misc{pith2026260805475,
  author       = {Pith},
  title        = {Pith review of: KV-Skill: Forging Expertise in the Model's Native Language},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3R3YEAI5}},
  note         = {Machine review of arXiv:2608.05475}
}
read the original abstract

Task knowledge is commonly stored either as text in the prompt or as an update to model weights. Text is modular but must be interpreted on every use, while weight adaptation makes the resulting capability difficult to load, remove, or share independently. We introduce KV-Skill, a design space of external factorized operators that a frozen language model reads through a lightweight interface. KV-Skill supports two complementary paths. Registration converts an authored text skill into a text-derived operator and trains a shared per-backbone interface. Reward learning develops a compact latent operator directly from task outcomes, with or without an authored skill. Neither path adds positions to the prompt. Across ten benchmarks and four backbones from three model families, converting text to a KV-Skill consistently makes the same procedural knowledge more effective. On Qwen3.5-4B LiveMath, registration reaches 77.2 accuracy, compared with 23.4 for the source text skill, 52.0 for SkillOpt, and 64.5 for SoftSkill. Under matched reward training and parameter budgets, KV-Skill gives the best result in seven of eight matched settings against soft prefixes, prefix tuning, and LoRA. A post-hoc rank analysis further shows that text-derived operators retain nearly all of their benefit with one task-aligned direction per injection layer, while matched random directions fail. Finally, one shared interface retains three independently loadable KV-Skills without measurable forgetting. These results show that task knowledge can be acquired from text or experience, compressed into an external operator, and deployed separately from the backbone. Code is available at: https://github.com/shawnzhg/KV-Skill

Figures

Figures reproduced from arXiv: 2608.05475 by the authors.

Figure 1
Figure 1. Overview of KV-Skill. (a) A KV-Skill is an external factorized associative operator, not text or an attention KV cache. Keys U (ℓ) s address slots; values W (ℓ) s return a query-dependent response. (b) Registration converts a text skill into per-layer operators M (ℓ) s , updating only the shared interface I reg b . (c) Reward learning optimizes a compact operator and task interface jointly, backbone frozen. The oper… view at source ↗
Figure 2
Figure 2. Reading a KV-Skill. The interface queries the loaded operator from the residual state and injects a gated response into the frozen backbone, separate from self-attention and without adding attention KV-cache entries. 3.3 Converting Text into a KV-Skill An authored text skill may contain useful knowl￾edge even when the model cannot reliably execute it from the prompt. We convert the document into a text-derived KV-Sk… view at source ↗
Figure 3
Figure 3. SearchQA system prompt template. The {skill_section} placeholder receives the skill document in the TEXT SKILL condition and is empty in all KV-Skill conditions. • When several nearby strings look similar, choose the one whose surrounding labels or layout best match the question. Exact Answer Discipline • Copy names, numbers, and dates exactly from the document whenever possible. • Prefer direct extraction over para… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Post-hoc rank and direction analysis. Left: spectral truncation preserves full-operator performance down to rank one while the registered interface remains fixed. Right: the correct rank-one direction preserves nearly all of the gain, whereas a magnitude-matched random…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 11 canonical work pages

  1. [1]

    Aho and Jeffrey D

    Alfred V. Aho and Jeffrey D. Ullman , title =. 1972

    1972

  2. [2]

    Publications Manual , year = "1983", publisher =

    1983

  3. [3]

    Chandra and Dexter C

    Ashok K. Chandra and Dexter C. Kozen and Larry J. Stockmeyer , year = "1981", title =. doi:10.1145/322234.322243

    work page arXiv 1981

  4. [4]

    Scalable training of

    Andrew, Galen and Gao, Jianfeng , booktitle=. Scalable training of

  5. [5]

    Dan Gusfield , title =. 1997

    1997

  6. [6]

    Tetreault , title =

    Mohammad Sadegh Rasooli and Joel R. Tetreault , title =. Computing Research Repository , volume =. 2015 , url =

    2015

  7. [7]

    A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =

    Ando, Rie Kubota and Zhang, Tong , Issn =. A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =. Journal of Machine Learning Research , Month = dec, Numpages =

  8. [8]

    arXiv preprint arXiv:2510.03215 , year=

    Cache-to-cache: Direct semantic communication between large language models , author=. arXiv preprint arXiv:2510.03215 , year=

    arXiv

Show all 47 references
  1. [9]

    ES-FoMo III: 3rd Workshop on Efficient Systems for Foundation Models , year=

    Cartridges: Lightweight and general-purpose long context representations via self-study , author=. ES-FoMo III: 3rd Workshop on Efficient Systems for Foundation Models , year=

  2. [10]

    arXiv preprint arXiv:2606.20333 , year=

    Softskill: Behavioral compression for contextual adaptation , author=. arXiv preprint arXiv:2606.20333 , year=

    arXiv

  3. [11]

    arXiv preprint arXiv:2605.23904 , year=

    Skillopt: Executive strategy for self-evolving agent skills , author=. arXiv preprint arXiv:2605.23904 , year=

    arXiv

  4. [12]

    Advances in neural information processing systems , volume=

    Parallelizing linear transformers with the delta rule over sequence length , author=. Advances in neural information processing systems , volume=

  5. [13]

    arXiv preprint arXiv:2605.22863 , year=

    Latent Cache Flow: Model-to-Model Communication Without Text , author=. arXiv preprint arXiv:2605.22863 , year=

    arXiv

  6. [14]

    Prefix-tuning: Optimizing continuous prompts for generation , author=. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers) , pages=

  7. [15]

    Proceedings of the 2021 conference on empirical methods in natural language processing , pages=

    The power of scale for parameter-efficient prompt tuning , author=. Proceedings of the 2021 conference on empirical methods in natural language processing , pages=

    2021

  8. [16]

    arXiv preprint arXiv:2305.16291 , year=

    Voyager: An open-ended embodied agent with large language models , author=. arXiv preprint arXiv:2305.16291 , year=

    arXiv

  9. [17]

    Advances in neural information processing systems , volume=

    Reflexion: Language agents with verbal reinforcement learning , author=. Advances in neural information processing systems , volume=

  10. [18]

    Advances in Neural Information Processing Systems , volume=

    Learning to compress prompts with gist tokens , author=. Advances in Neural Information Processing Systems , volume=

  11. [19]

    International conference on machine learning , pages=

    Transformers are rnns: Fast autoregressive transformers with linear attention , author=. International conference on machine learning , pages=. 2020 , organization=

    2020

  12. [20]

    International conference on machine learning , pages=

    Linear transformers are secretly fast weight programmers , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  13. [21]

    arXiv preprint arXiv:2308.10248 , year=

    Steering language models with activation engineering , author=. arXiv preprint arXiv:2308.10248 , year=

    arXiv

  14. [22]

    Findings of the Association for Computational Linguistics: EMNLP 2023 , pages=

    In-context learning creates task vectors , author=. Findings of the Association for Computational Linguistics: EMNLP 2023 , pages=

    2023

  15. [23]

    International conference on learning representations , volume=

    Function vectors in large language models , author=. International conference on learning representations , volume=

  16. [24]

    arXiv preprint arXiv:2507.08799 , year=

    KV Cache Steering for Controlling Frozen LLMs , author=. arXiv preprint arXiv:2507.08799 , year=

  17. [25]

    arXiv preprint arXiv:2410.10450 , year=

    Kblam: Knowledge base augmented language model , author=. arXiv preprint arXiv:2410.10450 , year=

    arXiv

  18. [26]

    arXiv preprint arXiv:2506.06105 , year=

    Text-to-lora: Instant transformer adaption , author=. arXiv preprint arXiv:2506.06105 , year=

    arXiv

  19. [27]

    arXiv preprint arXiv:2402.03300 , year=

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models , author=. arXiv preprint arXiv:2402.03300 , year=

    arXiv

  20. [28]

    arXiv preprint arXiv:2503.14476 , year=

    Dapo: An open-source llm reinforcement learning system at scale , author=. arXiv preprint arXiv:2503.14476 , year=

    arXiv

  21. [29]

    Approximating

    Schulman, John , year=. Approximating

  22. [30]

    On the Rollout-Training Mismatch in Modern

    Yao, Feng and others , year=. On the Rollout-Training Mismatch in Modern

  23. [31]

    arXiv preprint arXiv:2603.25158 , year=

    Trace2skill: Distill trajectory-local lessons into transferable agent skills , author=. arXiv preprint arXiv:2603.25158 , year=

    arXiv

  24. [32]

    differentiation

    Textgrad: Automatic" differentiation" via text , author=. arXiv preprint arXiv:2406.07496 , year=

    arXiv

  25. [33]

    First Workshop on Foundations of Reasoning in Language Models , year=

    Gepa: Reflective prompt evolution can outperform reinforcement learning , author=. First Workshop on Foundations of Reasoning in Language Models , year=

  26. [34]

    Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , pages=

    P-tuning: Prompt tuning can be comparable to fine-tuning across scales and tasks , author=. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , pages=

  27. [35]

    Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages=

    Adapting language models to compress contexts , author=. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages=

    2023

  28. [36]

    , author=

    Lora: Low-rank adaptation of large language models. , author=. Iclr , volume=

  29. [37]

    Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    Steering llm reasoning through bias-only adaptation , author=. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    2025

  30. [38]

    arXiv preprint arXiv:2509.06608 , year=

    Small vectors, big effects: A mechanistic study of rl-induced reasoning via steering vectors , author=. arXiv preprint arXiv:2509.06608 , year=

  31. [39]

    arXiv preprint arXiv:1704.05179 , year=

    Searchqa: A new q&a dataset augmented with context from a search engine , author=. arXiv preprint arXiv:1704.05179 , year=

    arXiv

  32. [40]

    2021 IEEE Winter Conference on Applications of Computer Vision (WACV) , pages=

    Docvqa: A dataset for vqa on document images , author=. 2021 IEEE Winter Conference on Applications of Computer Vision (WACV) , pages=. 2021 , organization=

    2021

  33. [41]

    Commonsenseqa: A question answering challenge targeting commonsense knowledge , author=. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) , pages=

    2019

  34. [42]

    Proceedings of the 2018 conference on empirical methods in natural language processing , pages=

    Can a suit of armor conduct electricity? a new dataset for open book question answering , author=. Proceedings of the 2018 conference on empirical methods in natural language processing , pages=

    2018

  35. [43]

    Advances in Neural Information Processing Systems , volume=

    Stark: Benchmarking llm retrieval on textual and relational knowledge bases , author=. Advances in Neural Information Processing Systems , volume=

  36. [44]

    arXiv preprint arXiv:2010.03768 , year=

    Alfworld: Aligning text and embodied environments for interactive learning , author=. arXiv preprint arXiv:2010.03768 , year=

    2010 arXiv

  37. [45]

    Advances in Neural Information Processing Systems , volume=

    Spreadsheetbench: Towards challenging real world spreadsheet manipulation , author=. Advances in Neural Information Processing Systems , volume=

  38. [46]

    arXiv preprint arXiv:2604.01754 , year=

    Livemathematicianbench: A live benchmark for mathematician-level reasoning with proof sketches , author=. arXiv preprint arXiv:2604.01754 , year=

  39. [47]

    arXiv preprint arXiv:2603.08655 , year=

    Officeqa pro: An enterprise benchmark for end-to-end grounded reasoning , author=. arXiv preprint arXiv:2603.08655 , year=

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.