Pith. sign in

REVIEW 4 major objections 5 minor 83 references

From Sweep to Seam: Interleaved Cross-Block Post-Training Quantization

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Interleaving the quantization schedule so each chunk-boundary pair is refined twice shrinks propagated activation error by a $\gamma^S$ factor and improves extreme low-bit perplexity.

desk verdict Neat schedule idea and an honest conditional bound, but the empirical seam attribution is confounded by re-rolls and extra compute, so the headline claim needs disentangling. read the letter →

arxiv 2608.09595 v1 pith:BQ2FPMNV submitted 2026-08-10 cs.AI

Achille Jacquemond , Yuma Ichikawa , Akira Sakai This is my paper · ORCID
classification cs.AI
keywords post-trainingquantizationcross-blocklargelanguagemodelslow-bitternaryschedulingerrorpropagationICBQ
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 paper argues that a scheduling change—not a new objective or architecture—can make extreme low-bit quantization of large language models substantially more stable. In cross-block post-training quantization, errors made at early block pairs propagate through the network; the paper's method, Interleaved Cross-Block Quantization (ICBQ), re-refines each boundary pair between consecutive chunks of blocks, so each "seam" is optimized twice. Under stated local contraction and smoothness assumptions, the paper proves a depth-wise bound in which seam revisits multiply the propagated error term by $\gamma^S$ while the additive residual stays bounded independently of depth. In experiments, ICBQ improves ternary-quantization perplexity relative to the matched sequential baseline on all seven main models, often turning divergent perplexity into finite values, and it also helps with 3-bit and 2-bit GPTQ.

What carries the argument

The carrying object is the chunk seam and its two-pass visit pattern. The outer driver quantizes blocks left-to-right in chunks of size $K$; at each chunk boundary it runs a two-block cross-block quantization (CBQ) refinement over a window that includes the last pair of the finished chunk and the first pair of the next chunk. Because the last pair of chunk $c$ is also the first pair of chunk $c+1$, that seam pair receives two CBQ passes. The paper isolates the seam effect in Lemma B.11 and shows the second pass reuses the same prefix, giving an extra contraction factor $\gamma$ on the propagated term. The memory cost is one stored layer-0 activation buffer, independent of depth and chunk size.

What would settle it

Log the pre- and post-pass midpoint residual per calibration sample at every CBQ call on a real model and compute the smallest uniform $\gamma$ satisfying $r_{\mathrm{out}} \le \gamma r_{\mathrm{in}} + \epsilon_{\mathrm{sub}}$. If any sample's post-pass residual exceeds the affine bound with $\gamma \ge 1$, or if no finite uniform $\gamma<1$ fits, the depth-wise bound of Theorem 4.1 does not hold.

Watch

Extended reading notes

Core claim

The central claim is that revisiting chunk seams—the boundary pairs between consecutive chunks of quantized Transformer blocks—converts a single left-to-right sweep into a schedule with repeated refinement at those seams, and this repetition alone contracts the propagated activation mismatch. The paper models the mismatch passed from one two-block window to the next, $m_i = \|\hat{X}_{i+1} - X^\star_{i+1}\|_F$, and shows each non-seam CBQ step obeys $m_i \le \tau m_{i-1} + (1+\gamma)\epsilon_{\mathrm{sub}}$, while a seam visit gives an extra factor $\gamma$: $m_i \le \gamma\tau m_{i-1} + (1+\gamma+\gamma^2)\epsilon_{\mathrm{sub}}$. Chaining these recurrences yields Theorem 4.1: with $S = \lceil L/K\rceil -1$ seams, the interleaved bound is $\gamma^S$ times the sequential propagated term plus a residual with a depth-independent constant. The paper reports that this schedule-level change lowers perplexity with ternary DBF on all seven evaluated models, and also applies to GPTQ at 3-bit and 2-bit precision.

Load-bearing premise

Every two-block CBQ subproblem must contract the per-sample midpoint residual by a uniform factor $\gamma \in (0,1)$ plus a fixed residual $\epsilon_{\mathrm{sub}}$; the paper notes rollback alone does not certify this and only measures an aggregate $\gamma$ on one model.

Editorial extensions

If this is right

  • Smaller chunk sizes create more seams and should improve the bound; the paper's experiments show perplexity improves monotonically or near-monotonically as $K$ goes from $L$ to 8, 4, and 2.
  • ICBQ transfers the schedule gain to GPTQ at W3 and W2g128, achieving best or tied-best perplexity on five of seven models.
  • The extra computation is only the re-refinement passes and one constant-size activation buffer, with no depth- or chunk-dependent memory growth.
  • The schedule keeps the local two-block objective, so it can be composed with stronger inner quantizers or local objectives rather than replacing them.

Reading between the lines

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

  • The same seam-revisit principle should apply to larger cross-block windows or other prefix-locality pipelines, since the theoretical argument only needs a contractive two-block subproblem and an exact re-roll.
  • A direct testable extension is to certify the uniform per-sample contraction constant by logging per-sample midpoint residuals; the paper leaves this for future work and only reports aggregate diagnostics.
  • If the schedule transfers beyond language models, as the paper leaves open, it would make extreme low-bit quantization a drop-in schedule change for other transformer families.
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 / 5 minor

Summary. The paper proposes Interleaved Cross-Block Quantization (ICBQ), a scheduling change to block-wise post-training quantization in which each chunk-boundary pair ('seam') is refined twice: once at the end of one chunk and again at the start of the next. The method keeps the local two-block CBQ objective and reuses calibration inputs, adding only a constant-size layer-0 activation buffer. Under stated Lipschitz, per-sample contraction, and exact re-roll assumptions, the authors derive an upper-bound comparison (Theorem 4.1 / Theorem B.5) in which each seam multiplies the propagated error term by an additional gamma factor, leading to an amortization gain prediction gamma^{-S}. Experiments on seven base models with ternary DBF and GPTQ W3/W2g128 report lower perplexity than a matched Sequential CBQ baseline, including cases where the baseline diverges, and an ablation varying chunk size K shows monotonic improvements as K decreases. The paper concludes that schedule design, specifically seam revisits, is the active ingredient.

Significance. If the empirical claims hold, ICBQ is a genuinely useful and cheap scheduling modification for extreme low-bit weight-only PTQ: it introduces no new objective, no architectural change, and only a constant-size memory buffer, while improving perplexity and zero-shot accuracy over a matched Sequential CBQ baseline across seven models and two inner quantizers. The paper is also commendable for shipping detailed algorithms, full proofs in the appendix, a numerical toy simulation, and a clear statement of the conditional nature of the theoretical result. However, the significance is bounded by two gaps: the theoretical contraction mechanism rests on Assumption B.2, which the authors explicitly say is not certified empirically in the per-sample uniform form required; and the empirical 'seam is the story' claim is confounded by concurrent changes in activation re-rolls and total refinement computation. Both points are addressable, but they are central to the paper's contribution as stated.

major comments (4)
  1. [§5.4, Tables 3–5; Algorithm 1 step 12; Algorithm 2 step 1] The empirical isolation of the seam effect is confounded. ICBQ differs from the Sequential CBQ baseline in at least three ways: (i) it performs periodic activation re-rolls from saved layer-0 inputs, which Sequential CBQ does not do; (ii) it performs more total pair-refinement updates (L-1+S vs. L-1, with measured wall-clock ratio ~1.21x in Table 5); and (iii) it revisits seam pairs. The chunk-size ablation in Table 3 varies K, which simultaneously changes the number of seams, the number of re-rolls, and the total refinement budget. The monotonic improvement with smaller K is therefore equally consistent with 'more re-rolls' or 'more compute' being the active ingredient rather than the seam revisit itself. Section 5.4's claim that 'the seam is the story' is not established. I request a control experiment that fixes re-roll frequency or total pair-update budget while varying only seam revisits, for example a Sequential CBQ variant with periodic re-rolls but no second seam pass, or an ICBQ variant with an equal number of extra non-seam pair refinements.
  2. [Assumption B.2, Eq. (21); Remark B.3; Appendix B.11, Eq. (74)] The theoretical centerpiece, the gamma^S factor in Theorem B.5 and Corollary B.8, depends on Assumption B.2: every actual two-block CBQ call must contract the per-sample midpoint residual uniformly by a constant gamma in (0,1), plus a fixed residual epsilon_sub. The paper itself states in Remark B.3 that rollback does not imply this certificate, and Appendix B.11 states that the empirically measured gamma is an aggregate diagnostic (mean 0.67, max 0.92) and 'is not, by itself, the uniform per-sample contraction constant required by Assumption B.2.' The proposed uniform estimator in Eq. (74) is not computed. Thus the theorem's prediction of geometric seam-amortized decay is not empirically supported in the uniform form needed; the theoretical claim is conditionally valid but its applicability to the reported systems remains open. I recommend either computing a uniform certificate on the actually visited finite sets (with an explicit epsilon_opt), or substantially reframing the theory section as a conditional bound and weakening the claim that the measured gamma diagnostics validate the amortization gain.
  3. [Theorem 4.1 and Table 5] The informal theorem statement says the comparison is made 'with the same CBQ optimizer budget,' but ICBQ performs S additional pair refinements over Sequential CBQ, and Table 5 reports a mean wall-clock increase of 1.21x. The formal theorem does not require equal total budget, but the informal statement is misleading and obscures the confound that part of ICBQ's gain could simply be additional optimization passes on the same objective. Please either remove 'same CBQ optimizer budget' from the informal statement, define it as per-call per-pair budget, or explicitly discuss the extra compute as part of the method's cost. The equal-budget claim currently contradicts the provided runtime data.
  4. [§5.3, Table 2] The GPTQ results show that ICBQ is not uniformly better than Sequential CBQ: at W3 on Mistral-7B ICBQ has PPL 8.86 versus 8.77 for Sequential CBQ, and there are several zero-shot task-level exceptions in Table 7. The text acknowledges this, but the framing in the abstract and conclusion that 'ICBQ reduces perplexity relative to the matched Sequential CBQ baseline' should be carefully qualified to the ternary DBF setting and to the aggregate perplexity comparisons. More importantly, the paper never compares against the original CBQ of Ding et al. or any other state-of-the-art block-wise PTQ method. Given that the claim is a schedule-level improvement, a comparison against the original multi-block CBQ window would establish practical relevance beyond the constructed matched baseline.
minor comments (5)
  1. [§3.2, Eq. (11) and Notation table] The definition of S(L,K) is inconsistent: Eq. (11) uses a floor expression while the notation table and Theorem 4.1 / Theorem B.5 use S(L,K)=ceil(L/K)-1. These differ when L is not divisible by K, and the proof of Lemma B.10 relies on the ceiling version. Please make the definitions consistent throughout.
  2. [§5.1 and Tables 1–4] All reported perplexity and zero-shot numbers appear to come from a single run with no error bars or multiple seeds. Given that the effects in Table 3 for K=8 vs K=4 are sometimes small (e.g., Llama-2-7B 13.07 vs 12.5), reporting variance across seeds or at least stating single-seed status would help assess the stability of the monotonic trend.
  3. [§5.4 header] The header 'The seam is the story' overstates what the ablation can show, given the confounds described in the first major comment. A more neutral title such as 'Chunk-size and schedule ablations' would better match the evidence.
  4. [Limitations (Section 7)] The Limitations paragraph mentions the block-locality and language-model scope, but it does not mention the unverified Assumption B.2 or the confounded seam ablation. Since these are the two main caveats on the paper's central claims, they should be acknowledged in the limitations.
  5. [Appendix C.3, Table 5] The wall-clock comparison reports 11 models, including four small models not present in the main seven-model tables (TinyLlama, Qwen3-0.6B, Gemma-2-2B, Gemma-3-1B). It would be helpful to state whether the main-text claims or only the runtime overhead claim use these additional models.

Circularity Check

1 steps flagged · score 3.0 of 10

Mild circularity: the γ-based amortization illustration is a rearrangement of a measured aggregate diagnostic; the core empirical comparison is independent.

  1. fitted input called prediction [Section 4, paragraph after Eq. (16); Appendix B.11]
    "For L= 32 and K= 4 (so S= 7 ), the measured aggregate ¯γ≈0.67 on Llama-2-7B (Appendix B.11) gives ¯γ−7 ≈16× , while a conservative aggregate maximum γ= 0.92 gives ≈1.8× ; the empirical range 1.08–14.5× in Section 5 lies between these diagnostics. ... The quantity ¯γreported here is an aggregate diagnostic ... It isnot, by itself, the uniform per-sample contraction constant required by Assumption B.2."

    The amortization gain γ^{-S} used to 'predict' the observed improvement is formed from γ measured on the same CBQ refinement calls whose effect the bound is meant to explain, and Appendix B.11 explicitly states that this measured γ is not the uniform per-sample constant required by Assumption B.2 and Theorem B.5. Consequently, the Section 4 numerical prediction (γ^{-7} ≈ 16× or ≈1.8×) is a transformation of a measured aggregate quantity, not an independent forecast; the agreement with the empirical range is a consistency check rather than confirmatory evidence. The raw PPL experiments remain independent of γ, so this is a mild, disclosed circularity rather than a forced fit.

full rationale

The core empirical contribution is self-contained: ICBQ and Sequential CBQ are compared under the same inner quantizer, objective, calibration set, and refinement budget, and the PPL and zero-shot results in Tables 1–4 do not depend on the theoretical γ. The theory (Theorem 4.1, formal Theorem B.5) is explicitly conditional on stated assumptions; Assumption B.2 provides a per-call midpoint contraction γ, and Lemma B.11 chains this assumption twice at each seam to obtain the γ^S factor. That is a normal conditional derivation rather than a self-definitional circle, since the per-call contraction is the input and the depth-level factor is the derived consequence. The only genuine circular element is the Section 4 illustration read as a prediction: the γ used there is measured from the same CBQ refinement calls, and Appendix B.11 explicitly denies that it is the uniform per-sample constant the theorem requires. The computed γ^{-S} gain is therefore a consistency check, not an independent forecast, and its agreement with the empirical range cannot provide the confirmatory force the presentation suggests. This is mild, because the core experimental comparison stands independently and the paper itself discloses the diagnostic limitation in Remark B.3 and Appendix B.11. Self-citations (MDBF, LPCD, OneComp, quasi-quantum annealing) appear in related-work and motivation contexts and are not load-bearing for the schedule comparison; no uniqueness result is imported from the authors' prior work. Overall score 3 reflects this mild input-measurement circularity in the theoretical illustration, while the central empirical claim retains independent content.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The theoretical result rests on three structural assumptions (B.1, B.2, B.4). The central one, Assumption B.2, postulates that every two-block CBQ call contracts each sample's midpoint error; this is the very mechanism the paper claims produces the improvements, and it is not independently verified. The empirical gamma used to illustrate the bound is measured from the same models, so the prediction of gain is a consistency check on the bound, not an external test.

free parameters (1)
  • gamma (per-sample midpoint contraction factor) = empirical mean approximately 0.67, max approximately 0.92 on Llama-2-7B (Appendix B.11); theorem requires a uniform…
    The bound's main gain gamma^S depends on this factor. The paper measures only an aggregate diagnostic, so the actual uniform gamma remains a free, unverified parameter.
assumptions (3)
  • domain assumption Teacher blocks are Lipschitz on all visited activation sets with a uniform constant rho (Assumption B.1)
    Used to propagate midpoint errors; rho is taken at most 1 in the standard stability regime, but the uniform bound over all calibration trajectories is not verified.
  • ad hoc to paper Each CBQ subproblem satisfies per-sample midpoint contraction with gamma in (0,1) and a fixed-point residual epsilon_sub (Assumption B.2, Eq. (21))
    This is the load-bearing certificate for the gamma^S factor. The paper notes in Remark B.3 that rollback does not imply it and that only aggregate gamma has been measured, so the axiom is assumed ad hoc for the derivation.
  • domain assumption Exact activation re-rolls with deterministic forward passes (Assumption B.4)
    Abstracts away floating-point non-associativity; needed for the prefix-invariance argument in Lemma B.11 and for the final-prefix identity in Lemma B.18.

how reviews work

0 comments
Cite this review

Pith. "Pith review of From Sweep to Seam: Interleaved Cross-Block Post-Training Quantization." pith.science (2026). https://pith.science/paper/BQ2FPMNV

@misc{pith2026260809595,
  author       = {Pith},
  title        = {Pith review of: From Sweep to Seam: Interleaved Cross-Block Post-Training Quantization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BQ2FPMNV}},
  note         = {Machine review of arXiv:2608.09595}
}
read the original abstract

Compressing large language models to two bits or fewer is increasingly feasible through block-wise post-training quantization; cross-block variants reconstruct neighboring Transformer blocks within a moving window. In the fixed two-block setting studied here, the matched sequential baseline moves this window through the network once, so errors introduced early in the sweep are not revisited. We propose Interleaved Cross-Block Quantization (ICBQ), a scheduling modification that revisits the boundary pair between consecutive chunks. Each seam pair is refined twice: first at the end of one chunk and again at the start of the next. The method retains the local two-block objective and reuses the calibration inputs of existing block-wise PTQ pipelines. Under stated local contraction and smoothness assumptions, we derive a depth-wise upper-bound comparison in which seam revisits multiply the propagated term while the residual remains bounded independently of depth. In the reported experiments, ICBQ reduces ternary-quantization perplexity relative to the matched Sequential CBQ baseline, yields finite perplexity in configurations where the baseline has severe degradation, and can also be used with 3-bit and 2-bit GPTQ.

Figures

Figures reproduced from arXiv: 2608.09595 by the authors.

Figure 1
Figure 1. ICBQ revisits chunk seams using the same calibration samples (L = 13, K = 4). Panel a. Sequential CBQ performs a single post-hoc sweep, optimizing each adjacent block pair once. Panel b. Interleaved CBQ inserts a short CBQ pass after each chunk of newly quantized blocks. Because the final window of chunk c is also the first window of chunk c+1, the seam pairs (3, 4), (7, 8), and (11, 12) are optimized twice, whereas… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

83 extracted references · 43 canonical work pages

  1. [1]

    2025 , url =

    Lee, Banseok and Kim, Dongkyu and You, Youngcheon and Kim, Youngmin , booktitle =. 2025 , url =

    2025

  2. [3]

    2024 , doi =

    Bulat, Adrian and Ouali, Yassine and Tzimiropoulos, Georgios , booktitle =. 2024 , doi =

    2024

  3. [4]

    2024 , doi =

    Liu, Zechun and Oguz, Barlas and Zhao, Changsheng and Chang, Ernie and Stock, Pierre and Mehdad, Yashar and Shi, Yangyang and Krishnamoorthi, Raghuraman and Chandra, Vikas , booktitle =. 2024 , doi =

    2024

  4. [5]

    2023 , eprint =

    Touvron, Hugo and Lavril, Thibaut and Izacard, Gautier and others , journal =. 2023 , eprint =

    work page 2023

  5. [6]

    2023 , eprint =

    Touvron, Hugo and Martin, Louis and Stone, Kevin and others , journal =. 2023 , eprint =

    work page 2023

  6. [7]

    Grattafiori, Aaron and Dubey, Abhimanyu and Jauhri, Abhinav and others , journal =. The. 2024 , eprint =

    work page 2024

  7. [8]

    and Sablayrolles, Alexandre and Mensch, Arthur and others , journal =

    Jiang, Albert Q. and Sablayrolles, Alexandre and Mensch, Arthur and others , journal =. Mistral. 2023 , eprint =

    work page 2023

  8. [9]

    2408.00118 , archivePrefix =

    arXiv preprint arXiv:2408.00118 , year =. 2408.00118 , archivePrefix =

    arXiv

Show all 83 references
  1. [10]

    2025 , eprint =

    Yang, An and Li, Anfeng and Yang, Baosong and others , journal =. 2025 , eprint =

    2025

  2. [11]

    2401.02385 , archivePrefix =

    Zhang, Peiyuan and Zeng, Guangtao and Wang, Tianduo and Lu, Wei , year =. 2401.02385 , archivePrefix =

    arXiv

  3. [12]

    2022 , eprint =

    Zhang, Susan and Roller, Stephen and Goyal, Naman and others , journal =. 2022 , eprint =

    2022

  4. [13]

    2024 , howpublished =

    2024

  5. [15]

    Proceedings of the 37th International Conference on Machine Learning , series =

    Up or Down? Adaptive Rounding for Post-Training Quantization , author =. Proceedings of the 37th International Conference on Machine Learning , series =. 2020 , url =

    2020

  6. [16]

    Li, Yuhang and Gong, Ruihao and Tan, Xu and Yang, Yang and Hu, Peng and Zhang, Qi and Yu, Fengwei and Wang, Wei and Gu, Shi , booktitle =

  7. [17]

    2022 , eprint =

    Frantar, Elias and Ashkboos, Saleh and Hoefler, Torsten and Alistarh, Dan , journal =. 2022 , eprint =

    2022

  8. [18]

    2023 , url =

    Xiao, Guangxuan and Lin, Ji and Seznec, Mickael and Wu, Hao and Demouth, Julien and Han, Song , booktitle =. 2023 , url =

    2023

  9. [19]

    2024 , url =

    Lin, Ji and Tang, Jiaming and Tang, Haotian and Yang, Shang and Chen, Wei-Ming and Wang, Wei-Chen and Xiao, Guangxuan and Dang, Xingyu and Gan, Chuang and Han, Song , booktitle =. 2024 , url =

    2024

  10. [20]

    Shao, Wenqi and Chen, Mengzhao and Zhang, Zhaoyang and Xu, Peng and Zhao, Lirui and Li, Zhiqian and Zhang, Kaipeng and Gao, Peng and Qiao, Yu and Luo, Ping , booktitle =

  11. [21]

    2024 , url =

    Liu, Jing and Gong, Ruihao and Wei, Xiuying and Dong, Zhiwei and Cai, Jianfei and Zhuang, Bohan , booktitle =. 2024 , url =

    2024

  12. [22]

    2023 , eprint =

    Yuan, Zhihang and Niu, Lin and Liu, Jiawei and others , journal =. 2023 , eprint =

    2023

  13. [23]

    , booktitle =

    Chee, Jerry and Cai, Yaohui and Kuleshov, Volodymyr and De Sa, Christopher M. , booktitle =. 2023 , doi =

    2023

  14. [24]

    and Li, Bo and Cameron, Pashmina and Jaggi, Martin and Alistarh, Dan and Hoefler, Torsten and Hensman, James , booktitle =

    Ashkboos, Saleh and Mohtashami, Amirkeivan and Croci, Maximilian L. and Li, Bo and Cameron, Pashmina and Jaggi, Martin and Alistarh, Dan and Hoefler, Torsten and Hensman, James , booktitle =. 2024 , doi =

    2024

  15. [25]

    Liu, Zechun and Zhao, Changsheng and Fedorov, Igor and Soran, Bilge and Choudhary, Dhruv and Krishnamoorthi, Raghuraman and Chandra, Vikas and Tian, Yuandong and Blankevoort, Tijmen , booktitle =

  16. [26]

    2025 , url =

    Ding, Xin and Liu, Xiaoyu and Tu, Zhijun and Zhang, Yun and Li, Wei and Hu, Jie and Chen, Hanting and Tang, Yehui and Xiong, Zhiwei and Yin, Baoqun and Wang, Yunhe , booktitle =. 2025 , url =

    2025

  17. [27]

    2023 , eprint =

    Wang, Hongyu and Ma, Shuming and Dong, Li and Huang, Shaohan and Wang, Haoran and Ma, Lingxiao and Yang, Fan and Wang, Ruiping and Wu, Yi and Wei, Furu , journal =. 2023 , eprint =

    2023

  18. [28]

    The Era of

    Ma, Shuming and Wang, Hongyu and Ma, Lingxiao and Wang, Lei and Wang, Wenhui and Huang, Shaohan and Dong, Li and Wang, Ruiping and Xue, Jilong and Wei, Furu , journal =. The Era of. 2024 , eprint =

    2024

  19. [29]

    2024 , doi =

    Xu, Yuzhuang and Han, Xu and Yang, Zonghan and Wang, Shuo and Zhu, Qingfu and Liu, Zhiyuan and Liu, Weidong and Che, Wanxiang , booktitle =. 2024 , doi =. 2402.11295 , archivePrefix =

    2024 arXiv

  20. [30]

    Transactions on Machine Learning Research , year =

    Addition Is Almost All You Need: Compressing Large Language Models with Double Binary Factorization , author =. Transactions on Machine Learning Research , year =

  21. [31]

    Advances in Neural Information Processing Systems , volume =

    Quantization Error Propagation: Revisiting Layer-Wise Post-Training Quantization , author =. Advances in Neural Information Processing Systems , volume =. 2025 , eprint =

    2025

  22. [32]

    2025 , eprint =

    Ichikawa, Yuma and Fujimoto, Yudai and Sakai, Akira , journal =. 2025 , eprint =

    2025

  23. [33]

    International Conference on Learning Representations , year =

    Optimization by Parallel Quasi-Quantum Annealing with Gradient-Based Sampling , author =. International Conference on Learning Representations , year =. 2409.02135 , archivePrefix =

    arXiv

  24. [34]

    2025 , eprint =

    Lin, Li and Wan, Xiaojun , journal =. 2025 , eprint =

    2025

  25. [35]

    2026 , eprint =

    Ichikawa, Yuma and Kimura, Keiji and Yoshida, Akihiro and Fujimoto, Yudai and Tokura, Hiroki and Arai, Yamato and Ishii, Yoshiyuki and Kawakami, Yusei and Shikada, Genki and Jacquemond, Achille and Fujisawa, Yoshihiko and Fujisawa, Katsuki and Honda, Takumi and Sakai, Akira , ...

    2026

  26. [36]

    Aligning

    Hendrycks, Dan and Burns, Collin and Basart, Steven and Critch, Andrew and Li, Jerry and Song, Dawn and Steinhardt, Jacob , booktitle =. Aligning. 2021 , url =

    2021

  27. [37]

    Journal of Machine Learning Research , volume =

    Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer , author =. Journal of Machine Learning Research , volume =. 2020 , url =

    2020

  28. [38]

    International Conference on Learning Representations , year =

    Pointer Sentinel Mixture Models , author =. International Conference on Learning Representations , year =

  29. [39]

    2021 , version =

    A Framework for Few-Shot Language Model Evaluation , author =. 2021 , version =. doi:10.5281/zenodo.5371629 , url =

    2021 doi

  30. [40]

    2020 , doi =

    Bisk, Yonatan and Zellers, Rowan and Le Bras, Ronan and Gao, Jianfeng and Choi, Yejin , journal =. 2020 , doi =

    2020

  31. [41]

    2019 , doi =

    Zellers, Rowan and Holtzman, Ari and Bisk, Yonatan and Farhadi, Ali and Choi, Yejin , booktitle =. 2019 , doi =

    2019

  32. [42]

    Think You Have Solved Question Answering?

    Clark, Peter and Cowhey, Isaac and Etzioni, Oren and Khot, Tushar and Sabharwal, Ashish and Schoenick, Carissa and Tafjord, Oyvind , journal =. Think You Have Solved Question Answering?. 2018 , eprint =

    2018

  33. [43]

    2021 , doi =

    Sakaguchi, Keisuke and Le Bras, Ronan and Bhagavatula, Chandra and Choi, Yejin , journal =. 2021 , doi =

    2021

  34. [44]

    2019 , doi =

    Clark, Christopher and Lee, Kenton and Chang, Ming-Wei and Kwiatkowski, Tom and Collins, Michael and Toutanova, Kristina , booktitle =. 2019 , doi =

    2019

  35. [45]

    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 , doi =

    2018

  36. [46]

    2018 , doi =

    High-Dimensional Probability: An Introduction with Applications in Data Science , author =. 2018 , doi =

    2018

  37. [47]

    Foundations and Trends in Machine Learning , volume =

    Distributed Optimization and Statistical Learning via the Alternating Direction Method of Multipliers , author =. Foundations and Trends in Machine Learning , volume =. 2011 , url =

    2011

  38. [48]

    The Annals of Mathematical Statistics , volume =

    On Information and Sufficiency , author =. The Annals of Mathematical Statistics , volume =. 1951 , doi =

    1951

  39. [49]

    2022 , url =

    Dettmers, Tim and Lewis, Mike and Belkada, Younes and Zettlemoyer, Luke , booktitle =. 2022 , url =

    2022

  40. [50]

    Quantization error propagation: Revisiting layer-wise post-training quantization

    Yamato Arai and Yuma Ichikawa. Quantization error propagation: Revisiting layer-wise post-training quantization. In Advances in Neural Information Processing Systems, volume 38, 2025. URL https://arxiv.org/abs/2504.09629

    2025

  41. [51]

    Croci, Bo Li, Pashmina Cameron, Martin Jaggi, Dan Alistarh, Torsten Hoefler, and James Hensman

    Saleh Ashkboos, Amirkeivan Mohtashami, Maximilian L. Croci, Bo Li, Pashmina Cameron, Martin Jaggi, Dan Alistarh, Torsten Hoefler, and James Hensman. QuaRot : Outlier-free 4-Bit inference in rotated LLM s. In Advances in Neural Information Processing Systems, volume 37, 2024. d...

    2024 doi

  42. [52]

    PIQA : Reasoning about physical commonsense in natural language

    Yonatan Bisk, Rowan Zellers, Ronan Le Bras, Jianfeng Gao, and Yejin Choi. PIQA : Reasoning about physical commonsense in natural language. Proceedings of the AAAI Conference on Artificial Intelligence, 34 0 (05): 0 7432--7439, 2020. doi:10.1609/aaai.v34i05.6239. URL https://oj...

    2020 doi

  43. [53]

    Addition is almost all you need: Compressing large language models with double binary factorization

    Vladim \'i r Bo z a and Vladim \'i r Macko. Addition is almost all you need: Compressing large language models with double binary factorization. Transactions on Machine Learning Research, 2026. ISSN 2835-8856. URL https://openreview.net/forum?id=k5kUKoewdQ

    2026

  44. [54]

    QBB : Quantization with binary bases for LLM s

    Adrian Bulat, Yassine Ouali, and Georgios Tzimiropoulos. QBB : Quantization with binary bases for LLM s. In Advances in Neural Information Processing Systems, volume 37, 2024. doi:10.52202/079017-0105. URL https://papers.nips.cc/paper_files/paper/2024/hash/05b69cc4c8ff6e24c5de...

    2024 doi

  45. [55]

    Jerry Chee, Yaohui Cai, Volodymyr Kuleshov, and Christopher M. De Sa. QuIP : 2-bit quantization of large language models with guarantees. In Advances in Neural Information Processing Systems, volume 36, 2023. doi:10.52202/075280-0196. URL https://papers.nips.cc/paper_files/pap...

    2023 doi

  46. [56]

    BoolQ : Exploring the surprising difficulty of natural yes/no questions

    Christopher Clark, Kenton Lee, Ming-Wei Chang, Tom Kwiatkowski, Michael Collins, and Kristina Toutanova. BoolQ : Exploring the surprising difficulty of natural yes/no questions. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computat...

    2019 doi

  47. [57]

    Think you have solved question answering? Try ARC , the AI2 reasoning challenge

    Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. Think you have solved question answering? Try ARC , the AI2 reasoning challenge. arXiv preprint arXiv:1803.05457, 2018. URL https://arxiv.org/abs/1803.05457

    2018 arXiv

  48. [58]

    CBQ : Cross-block quantization for large language models

    Xin Ding, Xiaoyu Liu, Zhijun Tu, Yun Zhang, Wei Li, Jie Hu, Hanting Chen, Yehui Tang, Zhiwei Xiong, Baoqun Yin, and Yunhe Wang. CBQ : Cross-block quantization for large language models. In International Conference on Learning Representations, 2025. URL https://openreview.net/f...

    2025

  49. [59]

    GPTQ : Accurate post-training quantization for generative pre-trained transformers

    Elias Frantar, Saleh Ashkboos, Torsten Hoefler, and Dan Alistarh. GPTQ : Accurate post-training quantization for generative pre-trained transformers. arXiv preprint arXiv:2210.17323, 2022. doi:10.48550/arXiv.2210.17323. URL https://arxiv.org/abs/2210.17323

    2022 doi arXiv

  50. [60]

    A framework for few-shot language model evaluation

    Leo Gao, Jonathan Tow, Stella Biderman, Sid Black, Anthony DiPofi, Charles Foster, Laurence Golding, Jeffrey Hsu, Kyle McDonell, Niklas Muennighoff, Jason Phang, Laria Reynolds, Eric Tang, Anish Thite, Ben Wang, Kevin Wang, and Andy Zou. A framework for few-shot language model...

    2021

  51. [61]

    Gemma 3 technical report

    Gemma Team , Aishwarya Kamath, Johan Ferret, Shreya Pathak, et al. Gemma 3 technical report. arXiv preprint arXiv:2503.19786, 2025. URL https://arxiv.org/abs/2503.19786

    2025 arXiv

  52. [62]

    The Llama 3 herd of models

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, et al. The Llama 3 herd of models. arXiv preprint arXiv:2407.21783, 2024. URL https://arxiv.org/abs/2407.21783

    2024 arXiv

  53. [63]

    Optimization by parallel quasi-quantum annealing with gradient-based sampling

    Yuma Ichikawa and Yamato Arai. Optimization by parallel quasi-quantum annealing with gradient-based sampling. In International Conference on Learning Representations, 2025. URL https://openreview.net/forum?id=9EfBeXaXf0

    2025

  54. [64]

    LPCD : Unified framework from layer-wise to submodule quantization

    Yuma Ichikawa, Yudai Fujimoto, and Akira Sakai. LPCD : Unified framework from layer-wise to submodule quantization. arXiv preprint arXiv:2512.01546, 2025 a . URL https://arxiv.org/abs/2512.01546

    2025

  55. [65]

    More than bits: Multi-envelope double binary factorization for extreme quantization

    Yuma Ichikawa, Yoshihiko Fujisawa, Yudai Fujimoto, Akira Sakai, and Katsuki Fujisawa. More than bits: Multi-envelope double binary factorization for extreme quantization. arXiv preprint arXiv:2512.24545, 2025 b . URL https://arxiv.org/abs/2512.24545

    2025

  56. [66]

    OneComp : One-line revolution for generative AI model compression

    Yuma Ichikawa, Keiji Kimura, Akihiro Yoshida, Yudai Fujimoto, Hiroki Tokura, Yamato Arai, Yoshiyuki Ishii, Yusei Kawakami, Genki Shikada, Achille Jacquemond, Yoshihiko Fujisawa, Katsuki Fujisawa, Takumi Honda, and Akira Sakai. OneComp : One-line revolution for generative AI mo...

    2026

  57. [67]

    Jiang, Alexandre Sablayrolles, Arthur Mensch, et al

    Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, et al. Mistral 7B . arXiv preprint arXiv:2310.06825, 2023. URL https://arxiv.org/abs/2310.06825

    2023 arXiv

  58. [68]

    LittleBit : Ultra low-bit quantization via latent factorization

    Banseok Lee, Dongkyu Kim, Youngcheon You, and Youngmin Kim. LittleBit : Ultra low-bit quantization via latent factorization. In Advances in Neural Information Processing Systems, volume 38, 2025. URL https://papers.nips.cc/paper_files/paper/2025/hash/a917c1d57088897beba47f96b4...

    2025

  59. [69]

    BRECQ : Pushing the limit of post-training quantization by block reconstruction

    Yuhang Li, Ruihao Gong, Xu Tan, Yang Yang, Peng Hu, Qi Zhang, Fengwei Yu, Wei Wang, and Shi Gu. BRECQ : Pushing the limit of post-training quantization by block reconstruction. In International Conference on Learning Representations, 2021

    2021

  60. [70]

    AWQ : Activation-aware weight quantization for on-device LLM compression and acceleration

    Ji Lin, Jiaming Tang, Haotian Tang, Shang Yang, Wei-Ming Chen, Wei-Chen Wang, Guangxuan Xiao, Xingyu Dang, Chuang Gan, and Song Han. AWQ : Activation-aware weight quantization for on-device LLM compression and acceleration. In Proceedings of Machine Learning and Systems, volum...

    2024

  61. [71]

    LoaQ : Layer-wise output approximation quantization

    Li Lin and Xiaojun Wan. LoaQ : Layer-wise output approximation quantization. arXiv preprint arXiv:2509.06297, 2025. URL https://arxiv.org/abs/2509.06297

    2025

  62. [72]

    QLLM : Accurate and efficient low-bitwidth quantization for large language models

    Jing Liu, Ruihao Gong, Xiuying Wei, Zhiwei Dong, Jianfei Cai, and Bohan Zhuang. QLLM : Accurate and efficient low-bitwidth quantization for large language models. In International Conference on Learning Representations, 2024 a . URL https://openreview.net/forum?id=FIplmUWdm3

    2024

  63. [73]

    LLM-QAT : Data-free quantization aware training for large language models

    Zechun Liu, Barlas Oguz, Changsheng Zhao, Ernie Chang, Pierre Stock, Yashar Mehdad, Yangyang Shi, Raghuraman Krishnamoorthi, and Vikas Chandra. LLM-QAT : Data-free quantization aware training for large language models. In Findings of the Association for Computational Linguisti...

    2024 doi

  64. [74]

    SpinQuant : LLM quantization with learned rotations

    Zechun Liu, Changsheng Zhao, Igor Fedorov, Bilge Soran, Dhruv Choudhary, Raghuraman Krishnamoorthi, Vikas Chandra, Yuandong Tian, and Tijmen Blankevoort. SpinQuant : LLM quantization with learned rotations. In International Conference on Learning Representations, 2025

    2025

  65. [75]

    The era of 1-Bit LLMs : All large language models are in 1.58 bits

    Shuming Ma, Hongyu Wang, Lingxiao Ma, Lei Wang, Wenhui Wang, Shaohan Huang, Li Dong, Ruiping Wang, Jilong Xue, and Furu Wei. The era of 1-Bit LLMs : All large language models are in 1.58 bits. arXiv preprint arXiv:2402.17764, 2024. URL https://arxiv.org/abs/2402.17764

    2024 arXiv

  66. [76]

    Llama 3.2 model card

    Meta Llama Team . Llama 3.2 model card. https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/MODEL_CARD.md, 2024. Accessed: 2026-05-01

    2024

  67. [77]

    Can a suit of armor conduct electricity? a new dataset for open book question answering

    Todor Mihaylov, Peter Clark, Tushar Khot, and Ashish Sabharwal. Can a suit of armor conduct electricity? a new dataset for open book question answering. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 2381--2391. Association for...

    2018 doi

  68. [78]

    WinoGrande : An adversarial winograd schema challenge at scale

    Keisuke Sakaguchi, Ronan Le Bras, Chandra Bhagavatula, and Yejin Choi. WinoGrande : An adversarial winograd schema challenge at scale. Communications of the ACM, 64 0 (9): 0 99--106, 2021. doi:10.1145/3474381

    2021 doi

  69. [79]

    OmniQuant : Omnidirectionally calibrated quantization for large language models

    Wenqi Shao, Mengzhao Chen, Zhaoyang Zhang, Peng Xu, Lirui Zhao, Zhiqian Li, Kaipeng Zhang, Peng Gao, Yu Qiao, and Ping Luo. OmniQuant : Omnidirectionally calibrated quantization for large language models. In International Conference on Learning Representations, 2024

    2024

  70. [80]

    Llama 2 : Open foundation and fine-tuned chat models

    Hugo Touvron, Louis Martin, Kevin Stone, et al. Llama 2 : Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288, 2023. URL https://arxiv.org/abs/2307.09288

    2023 arXiv

  71. [81]

    BitNet : Scaling 1-Bit transformers for large language models

    Hongyu Wang, Shuming Ma, Li Dong, Shaohan Huang, Haoran Wang, Lingxiao Ma, Fan Yang, Ruiping Wang, Yi Wu, and Furu Wei. BitNet : Scaling 1-Bit transformers for large language models. arXiv preprint arXiv:2310.11453, 2023. URL https://arxiv.org/abs/2310.11453

    2023 arXiv

  72. [82]

    SmoothQuant : Accurate and efficient post-training quantization for large language models

    Guangxuan Xiao, Ji Lin, Mickael Seznec, Hao Wu, Julien Demouth, and Song Han. SmoothQuant : Accurate and efficient post-training quantization for large language models. In Proceedings of the 40th International Conference on Machine Learning, volume 202 of Proceedings of Machin...

    2023

  73. [83]

    OneBit : Towards extremely low-bit large language models

    Yuzhuang Xu, Xu Han, Zonghan Yang, Shuo Wang, Qingfu Zhu, Zhiyuan Liu, Weidong Liu, and Wanxiang Che. OneBit : Towards extremely low-bit large language models. In Advances in Neural Information Processing Systems, volume 37, 2024. doi:10.52202/079017-2122. URL https://papers.n...

    2024 doi

  74. [84]

    Qwen3 technical report

    An Yang, Anfeng Li, Baosong Yang, et al. Qwen3 technical report. arXiv preprint arXiv:2505.09388, 2025. URL https://arxiv.org/abs/2505.09388

    2025 arXiv

  75. [85]

    HellaSwag : Can a machine really finish your sentence? In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4791--4800

    Rowan Zellers, Ari Holtzman, Yonatan Bisk, Ali Farhadi, and Yejin Choi. HellaSwag : Can a machine really finish your sentence? In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4791--4800. Association for Computational Linguistic...

    2019 doi

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.