Improving Prediction Certainty Estimation for Reliable Early Exiting via Null Space Projection

Duoqian Miao; Hongyun Zhang; Jianing He; Qi Zhang; Shufeng Hao; Yi Kun; Zhihua Wei

REVIEW 3 major objections 4 minor 28 references

Improving Prediction Certainty Estimation for Reliable Early Exiting via Null Space Projection

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a null-space-derived 'unknown class' logit yields more reliable early-exit decisions, reaching a 2.19x average speed-up on GLUE with negligible accuracy loss.

desk verdict A genuinely new exiting signal based on null-space projection; the mechanism needs a constant-l0 ablation to rule out a re-parameterized energy score. read the letter →

arxiv 2506.17249 v1 pith:LVVX6ARA submitted 2025-06-08 cs.LG cs.AI

Jianing He , Qi Zhang , Duoqian Miao , Yi Kun , Shufeng Hao , Hongyun Zhang , Zhihua Wei This is my paper

classification cs.LGcs.AI

keywords earlyexitingnullspaceprojectionpredictioncertaintyBERTGLUEinferenceaccelerationsoftmaxconfidenceclass-irrelevantinformation

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

Existing early-exiting systems for pre-trained language models decide when to stop inference based on logit-derived confidence, which the authors argue overestimates certainty because it ignores class-irrelevant information in the hidden features. The paper defines an NSP score---the fraction of a feature's norm lying in the null space of the classifier's weight matrix---and appends a scaled version of it to the logits as a virtual 'unknown' class. The resulting Certainty-Aware Probability (CAP) score, the softmax probability of that unknown class, is used as the exiting signal. On GLUE the method achieves an average 2.19x speed-up with negligible performance degradation, beating the previous state of the art by 28%.

What carries the argument

The central object is the orthogonal decomposition of the hidden feature with respect to the classifier's weight matrix: $x = x_W + x_{W^\perp}$, where $W$ is the column space and $W^\perp$ the null space of $W$. The NSP score, $\mathrm{NSP}(x) = \|x_{W^\perp}\| / \|x\|$, measures the proportion of class-irrelevant information; geometrically it is the cosine of the angle between the feature and its null-space projection. Scaled by $\alpha$, this scalar becomes a logit for a constructed UNK class, and the softmax probability of that class, $p_0$, is the CAP score used as the exiting signal. The mechanism turns the feature component the classifier cannot see into a measured 'unknownness' that tempers overconfident logit-based exits.

What would settle it

On a fixed multi-exit BERT, compute the area under the ROC curve of the NSP score, the softmax score, and the CAP score for the binary task of predicting whether an early-layer prediction is correct; if NSP and CAP do not beat the softmax score on most GLUE tasks, the claim that class-irrelevant information improves certainty estimation is falsified.

Watch

Extended reading notes

Core claim

The central claim is that prediction certainty is better estimated by combining class-relevant logits with the proportion of class-irrelevant information in the feature, measured by the NSP score. Formally, with classifier weight matrix $W$, the feature $x$ decomposes as $x = x_W + x_{W^\perp}$, and the NSP score is $\mathrm{NSP}(x) = \|x_{W^\perp}\| / \|x\|$. The CAP score is $p_0 = e^{\alpha\, \mathrm{NSP}(x)} / \bigl(\sum_{i=1}^{C} e^{l_i} + e^{\alpha\, \mathrm{NSP}(x)}\bigr)$, the softmax probability of a virtual UNK class. The paper claims this score yields more reliable exiting decisions, reducing premature exits while keeping delayed exits low, and that this reliability, not just higher confidence, explains the improved speed-accuracy trade-off.

Load-bearing premise

The method assumes that the null-space component of a hidden feature is class-irrelevant and that its relative size is a valid inverse proxy for prediction certainty, a claim the paper asserts without proof.

Editorial extensions

If this is right

At roughly 2.00x speed-up, the method holds GLUE accuracy close to BERT-base while averaging 2.19x acceleration across all six tasks.
Compared with the previous SOTA ConsistentEE, the method achieves a 28% higher speed-up at comparable accuracy.
Experiments show lower Premature Exiting Rate and comparable Delayed Exiting Rate, indicating the gain comes from avoiding exits on wrong early predictions.
The overhead is negligible: less than 1.21M FLOPs per layer and less than 0.03% additional parameters, and the method transfers to ALBERT.
The CAP score has higher Difficulty Inversion Scores than logit-based signals at most internal layers, indicating better certainty estimation.

Reading between the lines

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

Because the NSP score depends only on the trained classifier's weight matrix, it could be computed post hoc on any multi-exit network; the paper's experiments use jointly fine-tuned internal classifiers, so a fully post-hoc application is untested.
The per-task tuning of $\alpha$ and $\tau$ leaves open whether the 28% gain over ConsistentEE would survive a fixed-hyperparameter comparison across tasks.
The virtual UNK class gives CAP a natural interpretation as an open-set or out-of-distribution score, a test the paper does not run.
The orthogonal decomposition is defined for any linear classifier, so the CAP construction could transfer to vision or other architectures beyond BERT-style PLMs.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Short version: this is a modest but genuine contribution to early exiting, and the central idea is new enough to merit a referee's time. The CAP score, built from an NSP score and a virtual UNK logit, is a clean construction, and the paper does the right diagnostic work: DIS analysis and premature/delayed exit rates both favor CAP, and the computational overhead is tiny. I'd believe the reported gains are real, but not yet that the null-space mechanism is what drives them.

The paper's novelty is the specific use of the null-space projection of the hidden feature relative to the classifier weight matrix as a certainty signal. That is not present in the cited prior work. The geometric motivation is coherent, and the authors are careful to define the NSP score and the softmax extension. The empirical analyses in Section 6 are the paper's strongest asset: they show CAP tracks sample difficulty better than logit-based signals and reduces both premature and delayed exits. Code is available.

Soft spots, in order. First, the headline comparison mixes results from different papers and implementations; some baselines are from original papers, some from GPFEE, some from their own implementation. That is common in this literature but it does weaken the 28%-over-ConsistentEE claim. Second, α and the exit threshold τ are tuned per task, and the chosen values are not reported; no variance or significance is given, so the speed-up numbers are point estimates. Third, and most important for the mechanism: the paper never shows the distribution of NSP scores, nor does it ablate with a constant UNK logit. If NSP were near-constant, CAP would collapse to a monotone function of log-sum-exp, i.e. an energy score with a shifted threshold. The DIS results suggest NSP is not constant, because CAP beats E-LANG on DIS, but a constant-l0 ablation would settle it. This is a missing ablation, not a demonstrated flaw.

The circularity burden is low. The method is not fit to the target; only α and τ are tuned, which are standard. Self-citations are to the authors' own prior early-exiting papers and are not load-bearing.

Who this is for: anyone working on early exiting or confidence estimation for transformer inference. It deserves a serious referee, with a request for the constant-l0 ablation and per-task hyperparameter reporting.

Referee Report

3 major / 4 minor

Summary. The paper proposes a new early-exiting signal, the Certainty-Aware Probability (CAP) score, which augments the standard classifier logits with a scaled null-space-projection (NSP) score treated as a logit for a virtual 'UNK' class. The NSP score is defined as the norm of the feature component orthogonal to the classifier weight column space divided by the feature norm. The authors claim that previous logit-based signals overestimate certainty because they ignore class-irrelevant information, and that CAP yields more reliable exiting decisions. On GLUE, with BERT-base and ALBERT-base, the method is reported to achieve an average 2.19x speed-up over the backbone with negligible accuracy loss, outperforming ConsistentEE by 28% in speed, with additional DIS and error-rate analyses.

Significance. If the proposed mechanism is real, this is a useful plug-in heuristic: it adds negligible per-layer cost and does not require training the exiting signal. The paper is commendable for releasing code, reporting computational cost tables, and including DIS and premature/delayed exit-rate analyses. However, the central novelty is the claim that the null-space component carries sample-dependent certainty information; this claim is not yet empirically isolated. Because the method's advantage could alternatively stem from per-task hyperparameter tuning or from a monotone transform of existing logit-energy scores, the current evidence is suggestive rather than conclusive.

major comments (3)

[§3.3, Eq. (3); §4.2, Eq. (10)] The load-bearing assumption that the NSP score varies meaningfully across samples is not tested. For BERT hidden states with N=768 and C≤3, the classifier weight matrix has a null space of dimension at least 765; if feature energy is not strongly concentrated on the C class directions, NSP(x) is close to 1 and nearly constant. In that case CAP(x) becomes a monotone function of the log-sum-exp of the original logits (roughly a shifted energy score), and the reported gains could come from per-task α and τ tuning rather than from a null-space certainty signal. The paper reports no NSP distribution statistics (mean, standard deviation, range per layer) and no ablation that replaces the sample-dependent NSP with a constant logit. Please add these analyses to tie the observed speed-up to the proposed mechanism.
[§5.3-5.4, Table 2] The comparison against baselines is not controlled. Baseline numbers mix results taken from other papers (†, ∗) and from the authors' own implementation (‡), and the per-task values of α and the exiting threshold τ are not reported. The claim of a 28% average speed-up over ConsistentEE at 'negligible performance degradation' requires matched operating points and a common implementation environment; otherwise the advantage may reflect implementation or tuning differences. Please report α and τ for each task and rerun all compared baselines in the same codebase, or at least show that the reported points lie on the same speed-accuracy trade-off curve.
[§3.3] The statement that x_W⊥ is 'closely related to prediction certainty' and that a larger component 'interferes with classification and reduces prediction certainty' is an assertion, not a consequence of the orthogonal decomposition. The orthogonality W^T x_W⊥ = 0 only shows that the classifier is blind to that component; it does not establish that the component is detrimental or that its relative norm is an inverse certainty measure. The paper should validate this premise directly, for example by showing that NSP adds predictive information for the exit decision beyond the log-sum-exp of the original logits (e.g., through a partial-correlation or conditional-DIS analysis).

minor comments (4)

[Figure 3] The protocol for setting thresholds so that all methods operate at the same 4.00× speed-up ratio is not described; without this detail, the premature/delayed error-rate comparison is hard to interpret.
[Table 3] The DIS analysis is reported only on SST-2 and QNLI; the text claims that CAP 'consistently' outperforms baselines across different layers, which would be better supported by reporting DIS on all six GLUE tasks.
[§5.2] The choice to exclude FastBERT and other strong baselines is not justified in the paper; a sentence explaining the selection criteria would help the reader assess the completeness of the comparison.
[§5.3, Eq. (11)] The speed-up ratio is computed from the number of saved layers, not from measured wall-clock time; Table 5 provides an overhead check, but a statement that the two measures are consistent on the full GLUE suite would strengthen the efficiency claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the NSP and CAP definitions are proposed geometric quantities, and the reported gains are empirical comparisons rather than results assumed by construction.

full rationale

Walking the derivation chain, the paper defines NSP(x) = ||x_W_perp||/||x|| in Eq. (3) as a proposed geometric score, then defines l0 = alpha*NSP(x) in Eq. (8) and CAP(x) = p0 = exp(alpha*NSP(x)) / (sum_{i=1}^C exp(l_i) + exp(alpha*NSP(x))) in Eq. (10). These are definitions, not derivations that assume the target conclusion. The only fitted quantities are the scaling parameter alpha, selected per task from a small grid, and the exiting threshold tau, which are standard task-level hyperparameters in early exiting. The paper's claim that NSP encodes class-irrelevant certainty information is asserted as a hypothesis ('We hypothesize that combining class-relevant logits and the class-irrelevant NSP score could enhance prediction certainty estimation') and then tested empirically on GLUE; this is an unsupported premise or potential correctness risk, not circularity. The self-citations in the introduction and related work (He et al. 2024, 2025a, 2025b) are used only as background references for early exiting broadly and are not load-bearing for the definition of NSP or CAP, and no uniqueness theorem or prior result by the same authors is invoked to force the method. The absence of null-space distribution statistics or a constant-UNK-logit ablation, noted in the skeptic analysis, bears on whether the empirical gains are attributable to the null-space mechanism, but it is an experimental-support concern rather than a case where an equation reduces to its own input. No step in the paper's derivation chain fits any of the enumerated circularity patterns, so the honest finding is no significant circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 1 invented entities

The central claim rests on two tunable numbers (α and the threshold τ) and on the geometric assumption that the null-space projection measures class-irrelevant information. No new physical entities are introduced; the virtual UNK class is a modeling construct.

free parameters (2)

α (scaling parameter for NSP logit) = selected from {0.01, 0.1, 1.0, 10.0}; values not reported per task
Eq. (8) l0 = α·NSP(x); α aligns scale with original logits and is chosen per task, affecting CAP distribution and exiting decisions.
Exiting threshold τ = not reported; presumably tuned per task to achieve desired speed-up
Exiting occurs when CAP < τ; the threshold controls the speed-accuracy trade-off and is task-specific in all early-exit methods.

assumptions (4)

domain assumption W has full column rank so that W^T W is invertible in Eq. (6).
The projection formula x_W = W(W^T W)^{-1} W^T x requires invertibility; the paper does not discuss rank or regularization.
ad hoc to paper The null-space component x_perp is class-irrelevant and detrimental to classification certainty.
Core premise of Section 3.3; no proof, only geometric intuition and downstream empirical evidence.
ad hoc to paper The scaled NSP score can be treated as a logit for a virtual UNK class and combined with original logits via softmax.
The softmax extension in Eq. (9) is a modeling choice justified by semantic alignment in Section 4.3, not a mathematical necessity.
domain assumption Speed-up ratio in Eq. (11) assumes computational cost is proportional to the number of executed layers.
Used to report acceleration; reasonable for transformer encoders but ignores memory-bandwidth and fixed overheads.

invented entities (1)

Virtual UNK class
purpose: To convert the NSP score into a logit so softmax can produce a certainty-aware probability (CAP) that combines class-relevant and class-irrelevant information.
The UNK class is a construct appended to the classifier's output space. It has no existence outside the model and its usefulness is evaluated only through downstream accuracy, speed, DIS, and error-rate metrics.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Improving Prediction Certainty Estimation for Reliable Early Exiting via Null Space Projection." pith.science (2026). https://pith.science/paper/LVVX6ARA

@misc{pith2026250617249,
  author       = {Pith},
  title        = {Pith review of: Improving Prediction Certainty Estimation for Reliable Early Exiting via Null Space Projection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LVVX6ARA}},
  note         = {Machine review of arXiv:2506.17249}
}

read the original abstract

Early exiting has demonstrated great potential in accelerating the inference of pre-trained language models (PLMs) by enabling easy samples to exit at shallow layers, eliminating the need for executing deeper layers. However, existing early exiting methods primarily rely on class-relevant logits to formulate their exiting signals for estimating prediction certainty, neglecting the detrimental influence of class-irrelevant information in the features on prediction certainty. This leads to an overestimation of prediction certainty, causing premature exiting of samples with incorrect early predictions. To remedy this, we define an NSP score to estimate prediction certainty by considering the proportion of class-irrelevant information in the features. On this basis, we propose a novel early exiting method based on the Certainty-Aware Probability (CAP) score, which integrates insights from both logits and the NSP score to enhance prediction certainty estimation, thus enabling more reliable exiting decisions. The experimental results on the GLUE benchmark show that our method can achieve an average speed-up ratio of 2.19x across all tasks with negligible performance degradation, surpassing the state-of-the-art (SOTA) ConsistentEE by 28%, yielding a better trade-off between task performance and inference efficiency. The code is available at https://github.com/He-Jianing/NSP.git.

Figures

Figures reproduced from arXiv: 2506.17249 by the authors.

**Figure 1.** Method overview. Our method integrates the class-irrelevant NSP score with class-relevant logits to generate high-quality exiting [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗

**Figure 3.** Two types of error rates for exiting decisions using differ [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

**Figure 4.** Impact of α on the task performance under different speedup ratios for SST-2 and QNLI tasks. emphasis on the class-irrelevant NSP score compared to the class-relevant original logits. Overall, we observe that both excessively large and small values of α can impair the model acceleration under various speed-up ratios. This suggests an optimal trade-off between class-irrelevant and class-relevant information, which e… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

28 extracted references · 27 canonical work pages

[1]

E-LANG: Energy-Based Joint Inferencing of Super and Swift Language Models

Mohammad Akbari, Amin Banitalebi-Dehkordi, and Yong Zhang. E-lang: Energy-based joint inferencing of super and swift language models. arXiv preprint arXiv:2203.00748 , 2022

work page Pith review arXiv 2022
[2]

Palbert: Teaching albert to ponder

Nikita Balagansky and Daniil Gavrilov. Palbert: Teaching albert to ponder. Advances in Neural Information Processing Systems , 35:14002--14012, 2022

work page 2022
[3]

BERT: pre-training of deep bidirectional transformers for language understanding

Jacob Devlin, Ming - Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: pre-training of deep bidirectional transformers for language understanding. In NAACL-HLT (1) , pages 4171--4186. Association for Computational Linguistics, 2019

work page 2019
[4]

F-pabee: flexible-patience-based early exiting for single-label and multi-label text classification tasks

Xiangxiang Gao, Wei Zhu, Jiasheng Gao, and Congrui Yin. F-pabee: flexible-patience-based early exiting for single-label and multi-label text classification tasks. In ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages 1--5. IEEE, 2023

work page 2023
[5]

De$ ^ 3 $-bert: Distance-enhanced early exiting for BERT based on prototypical networks

Jianing He, Qi Zhang, Weiping Ding, Duoqian Miao, Jun Zhao, Liang Hu, and Longbing Cao. De$ ^ 3 $-bert: Distance-enhanced early exiting for BERT based on prototypical networks. CoRR , abs/2402.05948, 2024

work page arXiv 2024
[6]

COSEE: consistency-oriented signal-based early exiting via calibrated sample weighting mechanism

Jianing He, Qi Zhang, Hongyun Zhang, Xuanjing Huang, Usman Naseem, and Duoqian Miao. COSEE: consistency-oriented signal-based early exiting via calibrated sample weighting mechanism. In AAAI-25 , pages 24023--24031. AAAI Press, 2025

work page 2025
[7]

Two-stage early exiting from globality towards reliability

Jianing He, Qi Zhang, Hongyun Zhang, and Duoqian Miao. Two-stage early exiting from globality towards reliability. CAAI Transactions on Intelligence Technology , 2025

work page 2025
[8]

Early exit with disentangled representation and equiangular tight frame

Yixin Ji, Jikai Wang, Juntao Li, Qiang Chen, Wenliang Chen, and Min Zhang. Early exit with disentangled representation and equiangular tight frame. In Findings of the Association for Computational Linguistics: ACL 2023 , pages 14128--14142, 2023

work page 2023

Show all 28 references

[9]

ALBERT: A lite BERT for self-supervised learning of language representations

Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, and Radu Soricut. ALBERT: A lite BERT for self-supervised learning of language representations. In ICLR . OpenReview.net, 2020

2020
[10]

Cascadebert: Accelerating inference of pre-trained language models via calibrated complete models cascade

Lei Li, Yankai Lin, Deli Chen, Shuhuai Ren, Peng Li, Jie Zhou, and Xu Sun. Cascadebert: Accelerating inference of pre-trained language models via calibrated complete models cascade. In EMNLP (Findings) , pages 475--486. Association for Computational Linguistics, 2021

2021
[11]

A global past-future early exit method for accelerating inference of pre-trained language models

Kaiyuan Liao, Yi Zhang, Xuancheng Ren, Qi Su, Xu Sun, and Bin He. A global past-future early exit method for accelerating inference of pre-trained language models. In NAACL-HLT , pages 2013--2023. Association for Computational Linguistics, 2021

2013
[12]

Fastbert: a self-distilling BERT with adaptive inference time

Weijie Liu, Peng Zhou, Zhiruo Wang, Zhe Zhao, Haotang Deng, and Qi Ju. Fastbert: a self-distilling BERT with adaptive inference time. In ACL , pages 6035--6044. Association for Computational Linguistics, 2020

2020
[13]

Decoupled weight decay regularization

Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. In ICLR (Poster) . OpenReview.net, 2019

2019
[14]

Be3r: Bert based early-exit using expert routing

Sourab Mangrulkar, Ankith MS, and Vivek Sembium. Be3r: Bert based early-exit using expert routing. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages 3504--3512, 2022

2022
[15]

Roy Schwartz, Gabriel Stanovsky, Swabha Swayamdipta, Jesse Dodge, and Noah A. Smith. The right tool for the job: Matching model and instance complexities. In ACL , pages 6640--6651. Association for Computational Linguistics, 2020

2020
[16]

A simple hash-based early exiting approach for language understanding and generation

Tianxiang Sun, Xiangyang Liu, Wei Zhu, Zhichao Geng, Lingling Wu, Yilong He, Yuan Ni, Guotong Xie, Xuanjing Huang, and Xipeng Qiu. A simple hash-based early exiting approach for language understanding and generation. In ACL (Findings) , pages 2409--2421. Association for Comput...

2022
[17]

Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R. Bowman. GLUE: A multi-task benchmark and analysis platform for natural language understanding. In ICLR (Poster) . OpenReview.net, 2019

2019
[18]

Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, R \' e mi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger,...

2020
[19]

Deebert: Dynamic early exiting for accelerating BERT inference

Ji Xin, Raphael Tang, Jaejun Lee, Yaoliang Yu, and Jimmy Lin. Deebert: Dynamic early exiting for accelerating BERT inference. In ACL , pages 2246--2251. Association for Computational Linguistics, 2020

2020
[20]

Berxit: Early exiting for BERT with better fine-tuning and extension to regression

Ji Xin, Raphael Tang, Yaoliang Yu, and Jimmy Lin. Berxit: Early exiting for BERT with better fine-tuning and extension to regression. In EACL , pages 91--104. Association for Computational Linguistics, 2021

2021
[21]

Consistentee: A consistent and hardness-guided early exiting method for accelerating language models inference

Ziqian Zeng, Yihuai Hong, Hongliang Dai, Huiping Zhuang, and Cen Chen. Consistentee: A consistent and hardness-guided early exiting method for accelerating language models inference. In Thirty-Eighth AAAI Conference on Artificial Intelligence , pages 19506--19514. AAAI Press, 2024

2024
[22]

PCEE-BERT: accelerating BERT inference via patient and confident early exiting

Zhen Zhang, Wei Zhu, Jinfan Zhang, Peng Wang, Rize Jin, and Tae - Sun Chung. PCEE-BERT: accelerating BERT inference via patient and confident early exiting. In NAACL-HLT (Findings) , pages 327--338. Association for Computational Linguistics, 2022

2022
[23]

Leco: Improving early exiting via learned exits and comparison-based exiting mechanism

Jingfan Zhang, Ming Tan, Pengyu Dai, and Wei Zhu. Leco: Improving early exiting via learned exits and comparison-based exiting mechanism. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop) , pages 29...

2023
[24]

McAuley, Ke Xu, and Furu Wei

Wangchunshu Zhou, Canwen Xu, Tao Ge, Julian J. McAuley, Ke Xu, and Furu Wei. BERT loses patience: Fast and robust inference with early exit. In NeurIPS , 2020

2020
[25]

Gaml-bert: improving bert early exiting by gradient aligned mutual learning

Wei Zhu, Xiaoling Wang, Yuan Ni, and Guotong Xie. Gaml-bert: improving bert early exiting by gradient aligned mutual learning. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing , pages 3033--3044, 2021

2021
[26]

Badge: speeding up bert inference after deployment via block-wise bypasses and divergence-based early exiting

Wei Zhu, Peng Wang, Yuan Ni, Guotong Xie, and Xiaoling Wang. Badge: speeding up bert inference after deployment via block-wise bypasses and divergence-based early exiting. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 5: Ind...

2023
[27]

Leebert: Learned early exit for bert with cross-level optimization

Wei Zhu. Leebert: Learned early exit for bert with cross-level optimization. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers) , pages 29...

2021
[28]

write newline

" write newline "" before.all 'output.state := FUNCTION fin.entry add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence '...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

E-LANG: Energy-Based Joint Inferencing of Super and Swift Language Models

Mohammad Akbari, Amin Banitalebi-Dehkordi, and Yong Zhang. E-lang: Energy-based joint inferencing of super and swift language models. arXiv preprint arXiv:2203.00748 , 2022

work page Pith review arXiv 2022

[2] [2]

Palbert: Teaching albert to ponder

Nikita Balagansky and Daniil Gavrilov. Palbert: Teaching albert to ponder. Advances in Neural Information Processing Systems , 35:14002--14012, 2022

work page 2022

[3] [3]

BERT: pre-training of deep bidirectional transformers for language understanding

Jacob Devlin, Ming - Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: pre-training of deep bidirectional transformers for language understanding. In NAACL-HLT (1) , pages 4171--4186. Association for Computational Linguistics, 2019

work page 2019

[4] [4]

F-pabee: flexible-patience-based early exiting for single-label and multi-label text classification tasks

Xiangxiang Gao, Wei Zhu, Jiasheng Gao, and Congrui Yin. F-pabee: flexible-patience-based early exiting for single-label and multi-label text classification tasks. In ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages 1--5. IEEE, 2023

work page 2023

[5] [5]

De\( ^ 3 \)-bert: Distance-enhanced early exiting for BERT based on prototypical networks

Jianing He, Qi Zhang, Weiping Ding, Duoqian Miao, Jun Zhao, Liang Hu, and Longbing Cao. De\( ^ 3 \)-bert: Distance-enhanced early exiting for BERT based on prototypical networks. CoRR , abs/2402.05948, 2024

work page arXiv 2024

[6] [6]

COSEE: consistency-oriented signal-based early exiting via calibrated sample weighting mechanism

Jianing He, Qi Zhang, Hongyun Zhang, Xuanjing Huang, Usman Naseem, and Duoqian Miao. COSEE: consistency-oriented signal-based early exiting via calibrated sample weighting mechanism. In AAAI-25 , pages 24023--24031. AAAI Press, 2025

work page 2025

[7] [7]

Two-stage early exiting from globality towards reliability

Jianing He, Qi Zhang, Hongyun Zhang, and Duoqian Miao. Two-stage early exiting from globality towards reliability. CAAI Transactions on Intelligence Technology , 2025

work page 2025

[8] [8]

Early exit with disentangled representation and equiangular tight frame

Yixin Ji, Jikai Wang, Juntao Li, Qiang Chen, Wenliang Chen, and Min Zhang. Early exit with disentangled representation and equiangular tight frame. In Findings of the Association for Computational Linguistics: ACL 2023 , pages 14128--14142, 2023

work page 2023