Pith. sign in

REVIEW 4 major objections 7 minor 1 cited by

ERASE: EaRly bAckpropagation SchEdule for Faster Training of Modern Recommendation Systems

T0 review · 4 major / 7 minor · reviewed 2026-08-27 · deepseek-v4-flash

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

Pith's one-line read The paper establishes that Forward-Forward-style detachment can be used purely as a scheduling trick: each block's backward pass starts on a separate CUDA stream as soon as its forward pass ends, yielding up to 9.51% higher training…

desk verdict A smart scheduling idea undercut by a CTR experiment that never specifies the local losses, leaving the headline speedup ambiguous. read the letter →

arxiv 2608.18469 v1 pith:JOVQEQXL submitted 2026-08-19 cs.LG cs.AI

Ergan Shang , Flavio Sales Truzzi This is my paper · ORCID
classification cs.LGcs.AI
keywords earlybackwardpropagationForward-ForwarddetachmentCUDAstreamsGraphstrainingthroughputclick-through-ratemodelnormalizedentropygradientscheduling
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

The paper tries to establish that cutting a network into block-local subgraphs can speed up training without changing the mathematics of the gradients inside each subgraph. Its idea is that detaching a block's output with a stop-gradient makes that block's backward pass ready the moment its own forward pass and local objective finish, so the backward work can be launched early on a separate CUDA stream and overlapped with the forward pass of the next block. On a large click-through-rate model, ERASE reports p90 throughput gains up to 9.51% while keeping normalized entropy close to the baseline, and on a lightweight transformer it shows the overlap directly in an execution trace. The paper also marks the limit of the idea: a fused kernel that saturates the device leaves no spare capacity, so overlap helps only where kernels are small enough to leave the accelerator underutilized.

What carries the argument

The load-bearing mechanism is stop-gradient detachment used as a scheduling primitive. In the Forward-Forward-style graph, the output of block b is passed to block b+1 through a stop-gradient, so no gradient crosses block boundaries; with a local loss on the block's own output, the only inputs needed for that block's gradient are its own forward results and the local objective. ERASE launches each such subgraph's backward pass on a separate CUDA stream, using stream events to preserve the remaining dependencies, and in the non-blocking variant captures the affected subgraphs as CUDA Graphs to fix launch order across ranks. This is what turns idle capacity during the remainder of the forward pass into useful gradient work.

What would settle it

Run the blocking, FUP=True configuration on the same CTR model and measure its normalized-entropy gap over a full training run; if it is much larger than approximately 1.38%, the paper's headline claim that the best configuration keeps quality close to baseline is contradicted.

Watch

Extended reading notes

Core claim

The central claim is that reverse-mode training does not have to wait for the full forward pass before any backward work begins. Once a block's output is detached, its parameters and local loss form an independent subgraph, and ERASE dispatches that subgraph's backward pass as soon as its forward returns, accumulating gradients while later blocks still run forward. On a large click-through-rate model with six detached dense subarchitectures, the non-blocking configuration with CUDA Graphs raises p90 QPS to 198,145.28 (a 7.38% gain) with a normalized-entropy gap of about 1.38%, and the blocking configuration with FUP=True reaches 202,078.43 (a 9.51% gain). The paper's conclusion is that ERASE yields a 5-9% p90 QPS improvement with a small NE gap when cross-rank collective order is kept deterministic.

Load-bearing premise

The load-bearing premise is that the configuration with the highest throughput, which is not the one whose normalized-entropy gap was reported, keeps its NE gap close to the roughly 1.38% measured for the non-blocking CUDA Graphs row; if that gap grows materially with the blocking/FUP=True configuration, the promised combination of speed and quality is not established.

Editorial extensions

If this is right

  • Small-kernel workloads that leave the accelerator underutilized can convert otherwise idle capacity into training throughput without changing the loss or the per-subgraph gradients.
  • A device-saturating kernel, such as a fused multi-head attention backward, removes the free capacity that overlap depends on; the benefit is inherently workload-dependent.
  • Deterministic, rank-synchronized collective order is a practical requirement for the gain: blocking with FUP=True and non-blocking with CUDA Graphs both recover throughput, while plain non-blocking dispatch without fixed order does not.
  • On the tested click-through-rate model, ERASE improves p90 throughput by roughly 5-9% with a small normalized-entropy gap under deterministic collective order.

Reading between the lines

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

  • If the mechanism exploits spare capacity rather than reducing arithmetic, the gain should track the fraction of kernels that do not saturate the device; embedding-heavy and small-GEMM workloads are the natural place to look for larger effects.
  • The reported NE gap of about 1.38% belongs to the non-blocking + CUDA Graphs row; the paper does not report the NE gap for the best-throughput row, so the quality claim for that configuration is an extension from a nearby measurement, not a stated result.
  • A direct test of the saturation explanation: shrink hidden width or batch size on a small transformer and observe whether the overlap-induced speedup grows until the device-filling kernel becomes the bottleneck.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Request a human review

A listed scientist reviews the paper for a fee and the review publishes here regardless of verdict. See the reviewers or get listed.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper proposes ERASE, a scheduling scheme that uses Forward-Forward-style detachment to make each block's backward pass ready as soon as the block's forward pass finishes, then launches that backward pass early on a separate CUDA stream so it overlaps with subsequent forward work. The authors argue this improves accelerator utilization on small-kernel workloads, where conventional forward-then-backward scheduling leaves capacity idle. They demonstrate the intended overlap with a single-batch execution trace on a small transformer (NanoChat), note that a device-saturating fused attention kernel prevents further overlap, and then measure training throughput on a large-scale click-through-rate model with six detached dense subarchitectures. The reported result is a p90 QPS improvement of up to 9.51% over baseline with a 'small' normalized-entropy gap, although the only NE gap reported is about 1.38% for a configuration that achieves +7.38%.

Significance. If the empirical claims held, the scheduling idea would be a practical, inexpensive way to improve throughput in small-kernel training workloads. The conceptual derivation in Section 2 is clear and does not rely on fitted constants: the readiness of a block's backward pass follows directly from the stop-gradient construction, and the NanoChat execution trace is a useful sanity check that the intended overlap actually occurs. However, the manuscript currently does not specify the local objectives used for the detached CTR subarchitectures, leaving the meaning of the measured gradients ambiguous, and it does not report a quality metric for the best-throughput configuration. These two gaps weaken the central claim. The paper also does not provide code or a detailed reproducibility protocol, and the measurements are single-run values with no variance reporting.

major comments (4)
  1. [Section 3.2.1 / Eq. (2) / Section 5] The six detached subarchitectures are never assigned explicit local objectives. Eq. (2) defines the aggregate loss as L = L_main + Sum_i lambda_i ell_i, and the early backward for block i is only meaningful if ell_i is defined, but the CTR section never states what these ell_i are, how they are derived from the CTR label, what weights lambda_i were used, or which subgraphs receive them. Section 5 even says the scheme works 'without FF's goodness objective,' leaving unclear what gradient the early backward computes. As written, the +9.51% row could be consistent with the six subgraphs receiving zero gradient and therefore not being trained at all, making the throughput gain a comparison against an incompletely trained model. Please specify the per-subarchitecture losses or explain how the subgraph gradients are defined.
  2. [Table 1 / Sections 4 and 5] The best-throughput row (row 0, blocking + FUP=True, +9.51%) has no reported NE gap; the only quality measurement is approximately 1.38% for row 1 (non-blocking + CUDA Graphs, FUP=False). Since FUP explicitly changes which parameters participate in gradient synchronization, the NE behavior of row 0 cannot be assumed identical to row 1. The conclusion that ERASE gives '5-9% ... with a small NE gap' therefore extrapolates beyond the measured configuration. Please report the NE gap, and ideally the full NE curve, for row 0 and all other rows, or restrict the conclusion to the configuration for which quality was actually measured.
  3. [Table 1 / Section 3.2.2] The throughput numbers are single values with no repeated runs, confidence intervals, or variance across seeds or replicas. The p90 statistic is taken within one training run; no information is given on how many steps or how many replicas were used. The smallest gain (row 3, +0.37%) is within typical run-to-run noise of distributed training, so the ablation ordering in Table 1 cannot be distinguished from noise. Please run each configuration multiple times, report mean and standard deviation (or per-replica statistics), and state the number of steps used for the p90 computation.
  4. [Section 3.2.1] The CTR experiment description is too incomplete to reproduce: no model architecture (depth and width of the six detached subarchitectures), feature cardinality, dataset, optimizer, learning rate, batch size, data-parallel configuration on the eight H100 GPUs, or training duration is given. Without these details, it is impossible to determine whether the 5-9% speedup is tied to a particular kernel mix or model shape, and the broader claim about 'modern recommendation systems' is unsupported. Please add the missing experimental setup or clearly scope the claim.
minor comments (7)
  1. [Table 1] The dagger (†) on row 0 is never explained; if it flags the reported best configuration or refers to a footnote about quality, please state it explicitly. The row ordering (3, 2, 1, 0) is also confusing; reorder by gain or label rows consistently with the text.
  2. [Section 3.1] The MNIST sanity check is a single sentence with no hyperparameters, model size, or measurement protocol; please provide at least the setup and define what 'compute-matched backward time' means.
  3. [Section 4] The statement that 'FUP controls which parameters participate in gradient synchronization' should be made precise, since find_unused_parameters in DDP normally concerns gradient reduction for unused parameters; clarify the exact mechanism and why FUP=True reduces the parameters in the final aggregate backward.
  4. [Section 3.2.1] Please clarify why only four of the six detached subarchitectures receive separate CUDA streams; the scheduling effect, and hence the measured gain, may depend on which subgraphs are assigned streams.
  5. [Figure 2] The execution trace would be more informative with axis labels and a time scale; currently it reads as a schematic illustration.
  6. [Abstract / Section 5] The phrase 'normalized entropy close to the baseline' is vague; consider reporting the NE gap quantitatively in the abstract or defining 'close' in terms of the observed 1.38% gap.
  7. [Figure 3 / Section 3.2.2] The 'approximately 1.38%' NE gap is ambiguous between percentage points and relative change; please state the baseline NE value and define the gap in absolute and relative terms.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the throughput and NE numbers are direct measurements against a baseline, and the scheduling argument follows from stop-gradient semantics without fitted constants.

full rationale

ERASE's central claim is an empirical throughput measurement on a CTR model (Table 1) and an observed execution trace on NanoChat. No equation is fitted to produce the reported QPS gains, and no parameter is tuned to match the 9.51% number; the results are reports of measured runs. The scheduling rationale in Section 2.2 is a consequence of the definition of stop-gradient detachment: if a block's output is detached and a local loss is defined, its backward pass depends only on that block's forward and local objective. That is a graph-theoretic observation, not a prediction obtained by fitting. The paper's own Eq. (2) defines the aggregate loss, but the mechanism does not require Eq. (2) to be solved or inverted; it merely describes when each subgraph's backward becomes ready. Self-citations by the authors appear in the references and related-work discussion (e.g., diffusion debiasing, signed networks, ASI-bench), but none is load-bearing for the ERASE claim; the Forward-Forward mechanism is cited to Hinton [9], and CUDA Graphs to PyTorch and independent papers. The two concerns raised by a skeptical reading are genuine completeness issues, not circularity: Section 3.2.1 lists six detached subarchitectures without specifying the local objectives ℓ_i, leaving ambiguous what gradient the early backward computes; and the approximately 1.38% NE gap is reported only for the non-blocking + CUDA Graphs row (+7.38%), not for the best-throughput blocking + FUP=True row (+9.51%). These are unstated assumptions or missing measurements that could affect validity, but they do not make the derivation circular, because the reported gains are not constructed from the missing local objectives or from the omitted NE gap. The paper therefore does not reduce its central result to its own inputs by definition, fit, or self-citation chain.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the choice of detachment points and local loss weights, which are not reported. No new physical entities are introduced.

free parameters (2)
  • local loss coefficients lambda_i = not reported
    The CTR model uses L = L_main + sum lambda_i l_i with six detached subarchitectures, but the values of lambda_i are not given. The NE gap and throughput results depend on these coefficients.
  • number and placement of detachment points = 6 for CTR, 3 subgraphs for NanoChat
    Chosen by hand; no selection criterion or ablation is reported for varying the number of detachment points.
assumptions (3)
  • standard math Stop-gradient operator prevents gradient flow across block boundaries, so block-local backward is independent of downstream states.
    This is the definition of sg[...] in Section 2.1, used to claim that a block's backward pass becomes ready once its forward pass and local loss are complete.
  • domain assumption CUDA Graphs preserve launch order and determinism across ranks.
    Section 2.2 states this, but no verification or mechanism is provided. The deterministic behavior is load-bearing for the claimed throughput gains.
  • domain assumption The device has spare capacity during small-kernel forward passes that can be filled by concurrent backward work.
    Section 1 argues this, and Section 3.1 demonstrates one trace where overlap occurs, but it is not quantified generally and is explicitly limited by device-saturating kernels.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ERASE: EaRly bAckpropagation SchEdule for Faster Training of Modern Recommendation Systems." pith.science (2026). https://pith.science/paper/JOVQEQXL

@misc{pith2026260818469,
  author       = {Pith},
  title        = {Pith review of: ERASE: EaRly bAckpropagation SchEdule for Faster Training of Modern Recommendation Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JOVQEQXL}},
  note         = {Machine review of arXiv:2608.18469}
}
abstract

Lightweight proxy models enable rapid experimentation without repeatedly training frontier-scale systems, but their small kernels often leave modern accelerators underutilized. Conventional training compounds this inefficiency by scheduling the forward and backward passes as disjoint phases, so spare capacity in one cannot be filled by work from the other. We reinterpret the detachment mechanism of Forward-Forward (FF) as a scheduling primitive: given a local objective, detaching a block's output removes downstream gradient dependencies, making its backward pass ready when its forward pass finishes. ERASE launches each detached subgraph's backward pass early on a separate CUDA stream, overlapping it with subsequent forward work. Execution trace on a lightweight transformer demonstrates this overlap and its limit: a kernel that saturates the device leaves no capacity for concurrency. On a large-scale click-through-rate model, detaching six dense subarchitectures improves training throughput by up to $9.51\%$ while keeping normalized entropy close to the baseline.

Figures

Figures reproduced from arXiv: 2608.18469 by the authors.

Figure 1
Figure 1. Schematic of FF-style detachment. Each block receives a detached copy of the previous [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. NanoChat single-batch trace with two detachment points and three streams. Backward [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. NE gap between the non-blocking early-backward variant with CUDA Graphs (red curve) [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. LLM Evaluation on Unseen Questions: Contextual Multidimensional IRT Model

    cs.CL 2026-08 conditional novelty 5.0 of 10

    A contextual multidimensional IRT model improves within-scenario prediction of LLM performance on unseen questions, while cross-scenario transfer remains weak.

Reference graph

Works this paper leans on

29 extracted references · 21 canonical work pages · cited by 1 Pith paper

  1. [1]

    Lirank: Industrial large scale ranking models at linkedin

    Fedor Borisyuk, Mingzhou Zhou, Qingquan Song, Siyu Zhu, Birjodh Tiwana, Ganesh Parameswaran, Siddharth Dangi, Lars Hertel, Qiang Charles Xiao, Xiaochen Hou, et al. Lirank: Industrial large scale ranking models at linkedin. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 4804–4815, 2024

    work page 2024

  2. [2]

    Wide & deep learning for recommender systems

    Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, et al. Wide & deep learning for recommender systems. InProceedings of the 1st workshop on deep learning for recommender systems, pages 7–10, 2016

    2016

  3. [3]

    Deep neural networks for youtube recommenda- tions

    Paul Covington, Jay Adams, and Emre Sargin. Deep neural networks for youtube recommenda- tions. InProceedings of the 10th ACM conference on recommender systems, pages 191–198, 2016

    2016

  4. [4]

    Imagenet: A large- scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large- scale hierarchical image database. In2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009

    2009

  5. [5]

    The mnist database of handwritten digit images for machine learning research [best of the web].IEEE signal processing magazine, 29(6):141–142, 2012

    Li Deng. The mnist database of handwritten digit images for machine learning research [best of the web].IEEE signal processing magazine, 29(6):141–142, 2012

    2012

  6. [6]

    Boosting performance of iterative applications on gpus: Kernel batching with cuda graphs

    Jonah Ekelund, Stefano Markidis, and Ivy Peng. Boosting performance of iterative applications on gpus: Kernel batching with cuda graphs. In2025 33rd Euromicro International Conference on Parallel, Distributed, and Network-Based Processing (PDP), pages 70–77. IEEE, 2025

    work page 2025

  7. [7]

    Pygraph: Robust compiler support for cuda graphs in pytorch,

    Abhishek Ghosh, Ajay Nayak, Ashish Panwar, and Arkaprava Basu. Pygraph: Robust compiler support for cuda graphs in pytorch.arXiv preprint arXiv:2503.19779, 2025

    work page arXiv 2025

  8. [8]

    Training recommender systems at scale: Communication-efficient model and data parallelism

    Vipul Gupta, Dhruv Choudhary, Peter Tang, Xiaohan Wei, Xing Wang, Yuzhen Huang, Arun Kejariwal, Kannan Ramchandran, and Michael W Mahoney. Training recommender systems at scale: Communication-efficient model and data parallelism. InProceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, pages 2928–2936, 2021

    work page 2021

Show all 29 references
  1. [9]

    The forward-forward algorithm: Some preliminary investigations.arXiv preprint arXiv:2212.13345, 2022

    Geoffrey Hinton. The forward-forward algorithm: Some preliminary investigations.arXiv preprint arXiv:2212.13345, 2022

    2022 arXiv

  2. [10]

    Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    2020

  3. [11]

    autoresearch: Ai agents running research on single-gpu nanochat training automatically.https://github.com/karpathy/autoresearch, 2026

    Andrej Karpathy. autoresearch: Ai agents running research on single-gpu nanochat training automatically.https://github.com/karpathy/autoresearch, 2026

    2026

  4. [12]

    Federated online learning to rank with evolution strategies

    Eugene Kharitonov. Federated online learning to rank with evolution strategies. InProceedings of the Twelfth ACM International Conference on Web Search and Data Mining, pages 249–257, 2019

    2019

  5. [13]

    Noprop: Training neural networks without back-propagation or forward-propagation

    Qinyu Li, Yee Whye Teh, and Razvan Pascanu. Noprop: Training neural networks without back-propagation or forward-propagation. InConference on Lifelong Learning Agents, pages 525–544. PMLR, 2026

    2026

  6. [14]

    Long- horizon-terminal-bench: Testing the limits of agents on long-horizon terminal tasks with dense reward-based grading, 2026

    Zongxia Li, Zhongzhi Li, Yucheng Shi, Ruhan Wang, Junyao Yang, Zhichao Liu, Xiyang Wu, Anhao Li, Yue Yu, Ninghao Liu, Lichao Sun, Haotao Mi, and LeoweiLiang. Long- horizon-terminal-bench: Testing the limits of agents on long-horizon terminal tasks with dense reward-based gradi...

    2026 arXiv

  7. [15]

    Layer collabo- ration in the forward-forward algorithm

    Guy Lorberbom, Itai Gat, Yossi Adi, Alexander Schwing, and Tamir Hazan. Layer collabo- ration in the forward-forward algorithm. InProceedings of the AAAI conference on artificial intelligence, volume 38, pages 14141–14148, 2024

    2024

  8. [16]

    Understanding political polarization via jointly modeling users, connections and multimodal contents on heterogeneous graphs

    Hanjia Lyu and Jiebo Luo. Understanding political polarization via jointly modeling users, connections and multimodal contents on heterogeneous graphs. InProceedings of the 30th ACM international conference on multimedia, pages 4072–4082, 2022. 5

    2022

  9. [17]

    Software-hardware co-design for fast and scalable training of deep learning recommendation models

    Dheevatsa Mudigere, Yuchen Hao, Jianyu Huang, Zhihao Jia, Andrew Tulloch, Srinivas Sridha- ran, Xing Liu, Mustafa Ozdal, Jade Nie, Jongsoo Park, et al. Software-hardware co-design for fast and scalable training of deep learning recommendation models. InProceedings of the 49th ...

    2022

  10. [18]

    Humanity’s last exam.arXiv preprint arXiv:2501.14249, 2025

    Long Phan, Alice Gatti, Ziwen Han, Nathaniel Li, Josephina Hu, Hugh Zhang, Chen Bo Calvin Zhang, Mohamed Shaaban, John Ling, Sean Shi, et al. Humanity’s last exam.arXiv preprint arXiv:2501.14249, 2025

    2025 arXiv

  11. [19]

    Accelerating PyTorch with CUDA graphs

    PyTorch Team. Accelerating PyTorch with CUDA graphs. PyTorch Blog, https://pytorch. org/blog/accelerating-pytorch-with-cuda-graphs/, 2021. Accessed 2026-08-03

    2021

  12. [20]

    Predicting transcriptional outcomes of novel multigene perturbations with gears.Nature biotechnology, 42(6):927–935, 2024

    Yusuf Roohani, Kexin Huang, and Jure Leskovec. Predicting transcriptional outcomes of novel multigene perturbations with gears.Nature biotechnology, 42(6):927–935, 2024

    2024

  13. [21]

    Learning representations by back-propagating errors.nature, 323(6088):533–536, 1986

    David E Rumelhart, Geoffrey E Hinton, and Ronald J Williams. Learning representations by back-propagating errors.nature, 323(6088):533–536, 1986

    1986

  14. [22]

    David E Rumelhart, Geoffrey E Hinton, and Ronald J Williams. (1986) de rumelhart, ge hinton, and rj williams, learning internal representations by error propagation, parallel distributed pro- cessing: Explorations in the microstructures of cognition, vol. i, de rumelhart and j...

    1986

  15. [23]

    Ergan Shang, Yuting Wei, and Kathryn Roeder. Predicting the unseen: a diffusion-based debias- ing framework for transcriptional response prediction at single-cell resolution.Proceedings of the National Academy of Sciences, 122(52):e2525268122, 2025

    2025

  16. [24]

    Inference for balance in dynamic signed networks

    Ergan Shang, Yuan Zhang, and Weijing Tang. Inference for balance in dynamic signed networks. arXiv preprint arXiv:2606.08786, 2026

    2026 arXiv

  17. [25]

    Diffusionblocks: Block-wise neu- ral network training via diffusion interpretation

    Makoto Shing, Masanori Koyama, and Takuya Akiba. Diffusionblocks: Block-wise neu- ral network training via diffusion interpretation. InInternational Conference on Learning Representations, volume 2026, pages 95053–95074, 2026

    2026

  18. [26]

    Attention is all you need.Advances in neural information processing systems, 30, 2017

    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

    2017

  19. [27]

    Genetic convergence analysis of crispr perturbations deciphers gene functional similarity.bioRxiv, 2025

    Tianyu Zhang, Ergan Shang, and Kathryn Roeder. Genetic convergence analysis of crispr perturbations deciphers gene functional similarity.bioRxiv, 2025

    2025

  20. [28]

    Asi-bench: At the dawn of artificial superintelligence, 2026

    Junwei Zhou, Zhen Sun, Binyu Li, Jiangyu Zhou, Yuexi Pan, Hengyu Wang, Honghe Ren, Xiaohan Jia, Xueyang Zhou, Xiaoyu Cao, Yongchao Chen, Yuanning Feng, Junhao Wu, Cheng Zhang, Sijia Chen, Haoyu Xue, Chengsong You, Huan Wang, Koutian Wu, Peigan Gao, Jiakun Wu, Wenzhe Li, Ergan ...

    2026 arXiv

  21. [29]

    Rankmixer: Scaling up ranking models in industrial recommenders

    Jie Zhu, Zhifang Fan, Xiaoxie Zhu, Yuchen Jiang, Hangyu Wang, Xintian Han, Haoran Ding, Xinmin Wang, Wenlin Zhao, Zhen Gong, et al. Rankmixer: Scaling up ranking models in industrial recommenders. InProceedings of the 34th ACM International Conference on Information and Knowle...

    2025

Pith tools

Reviewed August 27, 2026 · model on record in the stance chip above.