Artificial Intelligence · arXiv:2611.16502 v1 · submitted 12 Nov 2026 · 41 references
SignRot: LLM Quantization with Massive Outlier-Aware Sign-Adjusted Rotation
The paper says
“While quantization compresses LLMs and accelerates inference, outlier activations impede efficient low-bit representation. In particular, massive outliers cause substantial performance degradation.”
more of the abstract
Although rotation-based methods that rely on randomized Hadamard transforms for online inference aim to mitigate this, they can produce fragmented distributions with empty bins when a massive outlier dominates a token's activation. This fragmentation severely undermines uniform quantization performance. Moreover, massive outliers concentrate in the down-projection input, where the preceding nonlinearity requires the rotation to be applied at inference time. This constraint prevents training-based approaches such as SpinQuant from learning the online rotation at this position, as a learned dense matrix cannot use the Fast Walsh-Hadamard Transform and would require a full matrix multiplication during inference. To address this, we propose SignRot (Sign-adjusted Rotation), which identifies massive outlier channels and constructs a sign adjustment vector that unifies the signs of the corresponding Hadamard row, collapsing the bimodal distribution into a unimodal profile amenable to quantization. Since the aligned outlier shifts the activation mean away from zero, we pair SignRot with asymmetric activation quantization, whose zero-point absorbs this shift. As a lightweight, drop-in improvement compatible with any Hadamard-based quantization method, SignRot requires only a single calibration forward pass and adds only element-wise sign flips at inference, without an additional matrix multiplication. Applying SignRot to QuaRot achieves a WikiText2 perplexity of 5.88 and an average zero-shot reasoning accuracy of 72.73, surpassing SpinQuant on the LLaMA3-70B model. Code is available at https://github.com/HYU-ARK-Lab/SignRot.
from the abstract
Pith says
Machine review · deepseek-v4-flash · 13 Nov 2026 · what was read was not recorded · not checked by a person
It can be wrong. Challenge this review or reply to it.
A clean, honest incremental improvement to Hadamard-based LLM quantization — well-motivated, well-tested, and modest in scope. This line is wrong
the referee pass did not run
How often objections from reviews like this one are right: not yet measured
Not checked: figures, code, what the 41 references say.
What would settle it
Reproduce the LLaMA3-70B GPTQ result (QuaRot+SignRot vs. QuaRot asymmetric) using the released code on the same hardware to confirm the WikiText2 perplexity drop from 6.16 to 5.88 and the average zero-shot accuracy improvement from 72.28 to 72.73 (Table 4).
from the skeptic's reading
How the review went
3 readings by deepseek-v4-flash · 13 Nov 2026
- Reader holds up
- Skeptic agrees
- Editor wrote the verdict above
Not run on this paper: the referee, the simulated authors' defence.
Pith did not check this review's objections against the paper.
The review in full
The review in brief, machine-written
This paper does one thing and does it well. It identifies a specific failure mode of random Hadamard transforms under massive outliers — bimodal fragmentation — and fixes it with a simple sign adjustment. The theoretical analysis of why the fragmentation occurs is clear and honestly caveated (the Gaussian-mixture approximation is approximate, and they say so). The fix is lightweight: one calibration pass, element-wise sign flips at inference, no extra matrix multiply. The experiments are thorough: multiple model families (LLaMA2/3, Mistral, Qwen3), two weight quantizers, multiple baselines (QuaRot, SpinQuant, DartQuant), latency benchmarks. The improvements are consistent but modest — 0.3–0.9 perplexity points, 1–3% accuracy — and the paper does not oversell them.
The soft spots are real but not hidden. The method assumes a single fixed dominant massive outlier channel per layer. This holds in early layers (where the impact is largest) but breaks in later layers, as the paper acknowledges in Appendix H. The gains there are zero. Also, the method requires asymmetric activation quantization to absorb the shifted mean, which costs a small amount of extra compute (though they measure it at <2% latency overhead). The theoretical derivation uses a Lindeberg condition that is not strictly satisfied in practice, but the key insight — that sign alternation from the Hadamard row causes the split — is empirically validated.
Who should read this? Anyone working on rotation-based LLM quantization, especially those using QuaRot or its derivatives. The method is drop-in and cheap enough to try even if the gains are moderate. The paper is well-written, the claims match the evidence, and the limitations are stated clearly. I would bring it to a reading group focused on efficient inference.
Recommendation: this deserves peer review. It is not a paradigm shift, but it is a solid, honest contribution that extends the known toolset. The experiments are reproducible (code provided), and the central claim is supported. I would suggest the authors add a clearer quantification of how much of the total improvement comes from asymmetric quantization vs. the sign adjustment itself — they partly address this in Table 4, but a direct ablation would help. Minor point: the activation clipping ratio is a free parameter; they report it in Table 13 but do not discuss sensitivity much. That said, the paper is in good shape and should go to a conference or journal in the ML systems / efficient NLP space.
The strongest objection
The paper explicitly acknowledges (Appendix H) that in later layers the dominant massive outlier channel varies per token, so SignRot does not fold those tokens. The reader's weakest assumption correctly notes this. However, the central claim is that SignRot improves overall quantization performance, and the empirical results (Table 1, Table 10) show gains despite this limitation. The first massive outlier layer, where the method is most effective, is shown to dominate quantization error, and SignRot reduces its KL divergence by 36–61% (Table 10). The overall perplexity and accuracy improvements are consistent across models. The limitation is honestly stated and does not invalidate the central claim; rather, it sets a clear scope. No hidden assumption or unaddressed flaw is present that would threaten the paper's main conclusion.
The reader identified the energy-dominance condition's breakdown in later layers as the weakest assumption. The paper acknowledges this limitation explicitly and does not claim effectiveness on those layers. The empirical evidence shows overall improvement, and the limitation is bounded. Therefore, this concern does not require a verdict change. The paper remains sound and well-supported.
Pith's reading of the paper (a machine's summary, not the authors')
Quantizing large language models to 4 bits is hard because some neurons produce unusually large activations called massive outliers. When a randomized Hadamard transform is used to spread these outliers across channels, a single huge value can create two separate clusters in the output distribution, with empty bins in between. This inflates the quantization step size and degrades accuracy, especially at the feed-forward down-projection where massive outliers concentrate and the transform must be applied online. SignRot fixes this problem by first identifying the outlier channel during a calibration forward pass. It then precomputes a sign-adjustment vector that flips the signs of selected Hadamard columns so that the row corresponding to the outlier channel becomes all positive. This turns the two clusters into a single shifted bell shape. Because the shifted mean is no longer centered at zero, SignRot also switches to asymmetric activation quantization, which uses a zero-point to absorb the shift. The sign flips are element-wise and add only about 1% to inference latency. Experiments on LLaMA2, LLaMA3, Mistral, and Qwen3 models show that adding SignRot to existing methods like QuaRot, SpinQuant, and DartQuant reduces perplexity by 0.2 to 0.9 points on WikiText-2 and raises average zero-shot reasoning accuracy by 1 to 3 percentage points, with the largest gains on the LLaMA3-70B model.
Core claim. SignRot improves quantization performance across diverse settings, including both dense and MoE architectures. On LLaMA3-70B with GPTQ, SignRot reduces the WikiText2 perplexity of QuaRot from 6.16 to 5.88 under the same asymmetric activation quantization, surpassing SpinQuant (6.38).
The premise it rests on. The method assumes that each massive outlier token has a single dominant channel that carries more than half the token's energy and that this channel is fixed across tokens (same cmax for the whole layer). This energy-dominance condition (Equation 3) is derived from a Gaussian-mixture approximation that requires the residual to satisfy a Lindeberg condition. The paper acknowledges in Appendix H that the assumption breaks in later layers where the largest channel varies per token, and that such tokens are not folded. The method's effectiveness on those layers is zero.
Check this review
Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.
How often reviews like this one are right
Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.
How often objections from reviews like this one are right: not yet measured
Reviews by people
This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.
The paper
Abstract
While quantization compresses LLMs and accelerates inference, outlier activations impede efficient low-bit representation. In particular, massive outliers cause substantial performance degradation. Although rotation-based methods that rely on randomized Hadamard transforms for online inference aim to mitigate this, they can produce fragmented distributions with empty bins when a massive outlier dominates a token's activation. This fragmentation severely undermines uniform quantization performance. Moreover, massive outliers concentrate in the down-projection input, where the preceding nonlinearity requires the rotation to be applied at inference time. This constraint prevents training-based approaches such as SpinQuant from learning the online rotation at this position, as a learned dense matrix cannot use the Fast Walsh-Hadamard Transform and would require a full matrix multiplication during inference. To address this, we propose SignRot (Sign-adjusted Rotation), which identifies massive outlier channels and constructs a sign adjustment vector that unifies the signs of the corresponding Hadamard row, collapsing the bimodal distribution into a unimodal profile amenable to quantization. Since the aligned outlier shifts the activation mean away from zero, we pair SignRot with asymmetric activation quantization, whose zero-point absorbs this shift. As a lightweight, drop-in improvement compatible with any Hadamard-based quantization method, SignRot requires only a single calibration forward pass and adds only element-wise sign flips at inference, without an additional matrix multiplication. Applying SignRot to QuaRot achieves a WikiText2 perplexity of 5.88 and an average zero-shot reasoning accuracy of 72.73, surpassing SpinQuant on the LLaMA3-70B model. Code is available at https://github.com/HYU-ARK-Lab/SignRot.
The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.
abstract compiled · 2026-11-13
The authors' main claim, as the classifier picked it out 0.89
“Applying SignRot to QuaRot achieves a WikiText2 perplexity of 5.88 and an average zero-shot reasoning accuracy of 72.73, surpassing SpinQuant on the LLaMA3-70B model.”
How the main claim is stated: asserted 1.00
What the abstract itself states, as probabilities: claims first 0.05 · states limitation 0.19 · human participants 0.02 · controlled comparison 0.85 · randomized assignment 0.03 · states quantitative result 0.99 · compares with prior results 0.98 · mentions code or data availability 0.99
The abstract, sentence by sentence
This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.
| Sentence | Role | The authors' own work | States the main conclusion | Strength |
|---|---|---|---|---|
| While quantization compresses LLMs and accelerates inference, outlier activations impede efficient low-bit representation. | background 1.00 | 0.06 | 0.03 | asserted 1.00 |
| In particular, massive outliers cause substantial performance degradation. | background 1.00 | 0.13 | 0.03 | asserted 1.00 |
| Although rotation-based methods that rely on randomized Hadamard transforms for online inference aim to mitigate this, they can produce fragmented distributions with empty bins when a massive outlier dominates a token's activation. | background 0.84 | 0.07 | 0.03 | hedged possibility 0.57 |
| This fragmentation severely undermines uniform quantization performance. | background 0.88 | 0.13 | 0.04 | asserted 1.00 |
| Moreover, massive outliers concentrate in the down-projection input, where the preceding nonlinearity requires the rotation to be applied at inference time. | background 0.91 | 0.31 | 0.03 | asserted 1.00 |
| This constraint prevents training-based approaches such as SpinQuant from learning the online rotation at this position, as a learned dense matrix cannot use the Fast Walsh-Hadamard Transform and would require a full matrix multiplication during inference. | background 0.68 | 0.14 | 0.03 | asserted 1.00 |
| To address this, we propose SignRot (Sign-adjusted Rotation), which identifies massive outlier channels and constructs a sign adjustment vector that unifies the signs of the corresponding Hadamard row, collapsing the bimodal distribution into a unimodal profile amenable to quantization. | contribution 0.65 | 0.92 | 0.43 | asserted 1.00 |
| Since the aligned outlier shifts the activation mean away from zero, we pair SignRot with asymmetric activation quantization, whose zero-point absorbs this shift. | method 0.91 | 0.73 | 0.06 | asserted 1.00 |
| As a lightweight, drop-in improvement compatible with any Hadamard-based quantization method, SignRot requires only a single calibration forward pass and adds only element-wise sign flips at inference, without an additional matrix multiplication. | contribution 0.59 | 0.79 | 0.13 | asserted 1.00 |
| Applying SignRot to QuaRot achieves a WikiText2 perplexity of 5.88 and an average zero-shot reasoning accuracy of 72.73, surpassing SpinQuant on the LLaMA3-70B model. | result 1.00 | 0.94 | 0.53 | asserted 1.00 |
| Code is available at https://github.com/HYU-ARK-Lab/SignRot. | contribution 0.57 | 0.44 | 0.02 | asserted 0.88 |
What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.83
What was examined, and what was not
Sources read
- arXiv listing abstract v1
Not examined
- full text
- figures and tables
- references and citing sentences
- relations to other papers
Levels run
- P abstract
Not run
- S sections
- T body sentences
- C citations
- R cross-paper relations
How the labels were made
Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-13: 1 request, 8,356 input tokens.
Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON
Figures (4)
Figures are the authors', reproduced from the paper.
References (41)
- [1] 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 10 inference in rotated llms.Advances in Neural Information Processing Systems, 37:100213– 100240, 2024 on Pith
- [2] Ben Berckmoes and Geert Molenberghs. Approximate central limit theorems.Journal of Theoretical Probability, 31(3):1590–1605, 2018 on Pith
- [3] Yonatan Bisk, Rowan Zellers, Ronan Le Bras, Jianfeng Gao, and Yejin Choi. Piqa: Reasoning about physical commonsense in natural language. InThirty-Fourth AAAI Conference on Artificial Intelligence, 2020 on Pith
- [4] Jerry Chee, Yaohui Cai, V olodymyr Kuleshov, and Christopher M De Sa. Quip: 2-bit quantiza- tion of large language models with guarantees.Advances in Neural Information Processing Systems, 36:4396–4429, 2023 on Pith
- [5] 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 on PithPith's reviewarXiv
- [6] Samuel Rhys Cox and Wei Tsang Ooi. Conversational interactions with npcs in llm-driven gaming: Guidelines from a content analysis of player feedback. InInternational Workshop on Chatbot Research and Design, pages 167–184. Springer, 2023 on Pith
- [7] Harald Cramér. Über eine Eigenschaft der normalen Verteilungsfunktion.Mathematische Zeitschrift, 41(1):405–414, 1936 on Pith
- [8] Tim Dettmers, Mike Lewis, Younes Belkada, and Luke Zettlemoyer. Gpt3. int8 (): 8-bit matrix multiplication for transformers at scale.Advances in neural information processing systems, 35:30318–30332, 2022 on Pith
- [9] Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. The llama 3 herd of models.arXiv preprint arXiv:2407.21783, 2024 on PithPith's reviewarXiv
- [10] Fino and Algazi. Unified matrix treatment of the fast walsh-hadamard transform.IEEE Transactions on Computers, 100(11):1142–1146, 1976 on Pith
- [11] Elias Frantar, Saleh Ashkboos, Torsten Hoefler, and Dan Alistarh. OPTQ: Accurate quantization for generative pre-trained transformers. InThe Eleventh International Conference on Learning Representations, 2023 on Pith
- [12] Coleman Hooper, Sehoon Kim, Hiva Mohammadzadeh, Michael W Mahoney, Yakun S Shao, Kurt Keutzer, and Amir Gholami. Kvquant: Towards 10 million context length llm inference with kv cache quantization.Advances in Neural Information Processing Systems, 37:1270–1303, 2024 on Pith
- [13] Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas ... on Pith
- [14] Jemin Lee, Sihyeong Park, Jinse Kwon, Jihun Oh, and Yongin Kwon. Exploring the trade-offs: Quantization methods, task difficulty, and model size in large language models from edge to giant. InProceedings of the Thirty-Fourth International Joint Conference on Artificial Intelli... on Pith
- [15] Jun Li, Fuxin Li, and Sinisa Todorovic. Efficient riemannian optimization on the stiefel manifold via the cayley transform. InInternational Conference on Learning Representations, 2020 on Pith
- [16] Zhuohan Li, Eric Wallace, Sheng Shen, Kevin Lin, Kurt Keutzer, Dan Klein, and Joseph E. Gonzalez. Train large, then compress: Rethinking model size for efficient training and inference of transformers. InProceedings of the 37th International Conference on Machine Learning, 2020. 11 on Pith
- [17] Haokun Lin, Haobo Xu, Yichen Wu, Jingzhi Cui, Yingtao Zhang, Linzhan Mou, Linqi Song, Zhenan Sun, and Ying Wei. Duquant: Distributing outliers via dual transformation makes stronger quantized llms.Advances in Neural Information Processing Systems, 37:87766–87800, 2024 on Pith
- [18] Ruikang Liu, Haoli Bai, Haokun Lin, Yuening Li, Han Gao, Zhengzhuo Xu, Lu Hou, Jun Yao, and Chun Yuan. Intactkv: Improving large language model quantization by keeping pivot tokens intact. InFindings of the Association for Computational Linguistics: ACL 2024, 2024 on Pith
- [19] 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. InThe Thirteenth International Conference on Learning Representations, 2025 on Pith
- [20] Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer sentinel mixture models. InInternational Conference on Learning Representations, 2016 on Pith
show 21 more
- [21] NVIDIA. Nvidia cutlass library, 2023 on Pith
- [22] Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style, high-performance deep learning library.Advances in neural information processing systems, 32, 2019 on Pith
- [23] Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. Language models are unsupervised multitask learners.OpenAI blog, 1(8):9, 2019 on Pith
- [24] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer.The Journal of Machine Learning Research, 21(1):5485–5551, 2020 on Pith
- [25] Baptiste Roziere, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat, Xiaoqing Ellen Tan, Yossi Adi, Jingyu Liu, Romain Sauvestre, Tal Remez, et al. Code llama: Open foundation models for code.arXiv preprint arXiv:2308.12950, 2023 on PithPith's reviewarXiv
- [26] Keisuke Sakaguchi, Ronan Le Bras, Chandra Bhagavatula, and Yejin Choi. Winogrande: An adversarial winograd schema challenge at scale.Communications of the ACM, 64(9):99–106, 2021 on Pith
- [27] Mark F. Schilling, Ann E. Watkins, and William Watkins. Is human height bimodal?The American Statistician, 56(3):223–229, 2002 on Pith
- [28] Wenqi Shao, Mengzhao Chen, Zhaoyang Zhang, Peng Xu, Lirui Zhao, Zhiqian Li, Kaipeng Zhang, Peng Gao, Yu Qiao, and Ping Luo. Omniquant: Omnidirectionally calibrated quantiza- tion for large language models.arXiv preprint arXiv:2308.13137, 2023 on PithPith's reviewarXiv
- [29] Yuantian Shao, Yuanteng Chen, Peisong Wang, Jianlin Yu, Jing Lin, Yiwu Yao, Zhihui Wei, and Jian Cheng. Dartquant: Efficient rotational distribution calibration for LLM quantization. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025 on Pith
- [30] Mingjie Sun, Xinlei Chen, J. Zico Kolter, and Zhuang Liu. Massive activations in large language models.arXiv preprint arXiv:2402.17762, 2024 on PithPith's reviewarXiv
- [31] Yuxuan Sun, Ruikang Liu, Haoli Bai, Han Bao, Kang Zhao, Yuening Li, Jiaxin Hu, Xianzhi Yu, Lu Hou, Chun Yuan, et al. Flatquant: Flatness matters for llm quantization.arXiv preprint arXiv:2410.09426, 2024 on PithPith's reviewarXiv
- [32] Arun James Thirunavukarasu, Darren Shu Jeng Ting, Kabilan Elangovan, Laura Gutierrez, Ting Fang Tan, and Daniel Shu Wei Ting. Large language models in medicine.Nature medicine, 29(8):1930–1940, 2023 on Pith
- [33] Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. Llama 2: Open foundation and fine-tuned chat models.arXiv preprint arXiv:2307.09288, 2023. 12 on PithPith's reviewarXiv
- [34] Albert Tseng, Jerry Chee, Qingyao Sun, V olodymyr Kuleshov, and Christopher De Sa. Quip#: Even better llm quantization with hadamard incoherence and lattice codebooks.arXiv preprint arXiv:2402.04396, 2024 on PithPith's reviewarXiv
- [35] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in Neural Information Processing Systems, 30, 2017 on Pith
- [36] T Wolf. Huggingface’s transformers: State-of-the-art natural language processing.arXiv preprint arXiv:1910.03771, 2019 on PithPith's reviewarXiv
- [37] Jingyang Xiang and Saiqian Zhang. Dfrot: Achieving outlier-free and massive activation-free for rotated llms with refined rotation.arXiv preprint arXiv:2412.00648, 2024 on PithPith's reviewarXiv
- [38] 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 Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett, ed... on Pith
- [39] An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jia... on Pith
- [40] Zihao Ye. FlashInfer: Kernel Library for LLM Serving. https://github.com/ flashinfer-ai/flashinfer, 2023 on Pith
- [41] Rowan Zellers, Ari Holtzman, Yonatan Bisk, Ali Farhadi, and Yejin Choi. Hellaswag: Can a machine really finish your sentence? InProceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4791–4800, 2019. A Derivation of the bimodality thresh... on Pith
Challenges and discussion
Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.
Come back by email
A short list of the week's newly reviewed papers. Nothing is sent until you confirm from your inbox, and every email has a link that stops it. No account.
No comments yet.
Sign in to comment