Pith. sign in

REVIEW 3 major objections 4 minor 38 references

Penelope: Localized Latent Recurrence for Efficient Structured Reasoning

T0 review · 3 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read Penelope localizes latent reasoning to a narrow decoder interval, matching full-decoder accuracy while roughly halving inference latency on structured reasoning tasks.

desk verdict Genuine architectural contribution with careful experiments, but the headline latency gain is oversold; the matched comparison still supports a real efficiency win. read the letter →

arxiv 2607.25915 v1 pith:N7G2QCY5 submitted 2026-07-28 cs.AI

Yutong Chen , Shouqian Shi , Xinran Liu , Haochen Wang , Jiaying Wang , Tianxing Xu , Yuanxi Wang , Zirui Ding This is my paper · ORCID
classification cs.AI
keywords latentreasoninglocalizedrecurrencedecoder-onlytransformerschain-of-thoughtdistillationinferencelatencystructuredtest-timecomputerecurrentmemory
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 tries to establish that a pretrained decoder-only language model can do latent, hidden-state reasoning by repeatedly executing only a narrow slice of its layers rather than the whole network. Penelope builds a fixed problem-state memory with one pass through the lower layers, then loops a late 5-layer interval K times to refine that memory, and only then generates the answer. Across three structured-reasoning tasks, this localized loop matches the accuracy of full-decoder latent reasoning (e.g., 52.25% vs 52.79% exact match on Deep ListOps) while cutting sequential decoder-layer applications from 160 to 46 and measured latency from 188 ms to 100 ms. The broader point is that additional reasoning compute can be allocated cheaply and locally instead of scaling parameters or generating long visible traces.

What carries the argument

The localized latent interface: a decomposition of the decoder into a lower prefix computed once, a recurrent interval F[ℓ_s:ℓ_e) of width r=5, and a final answer pass. The prefix produces a problem-conditioned boundary memory M₀ (m=8 anchors) and readout state R₀ (q=2 registers); then, K times, the interval is applied to the memory plus step- and group-embeddings, updating it via a time-modulated GRU with residual adapter and gated interpolation (M_t = M_{t-1} + 0.5α_t(fM_t − M_{t-1})), while a readout GRU maintains answer-accessible state. A prompt-side KV cache built once allows the loop to attend to fixed context without re-running the prefix; the cost formula C = L + (K+1)r makes the ma

What would settle it

Run Penelope on a structured-reasoning task whose minimal state size provably exceeds 8 values (e.g., a multi-counter or multi-hop graph task), sweep the number of memory anchors m from 1 to 32, and compare against a full-decoder latent baseline: if accuracy is substantially lower at m=8 and does not recover when m increases, then the fixed interface—and with it the localization claim—does not transfer to tasks requiring richer state.

Watch

Extended reading notes

Core claim

The central claim is that localization is computationally sufficient: a one-time lower-prefix pass creates a persistent boundary memory (8 latent slots and 2 readout registers), and iterative refinement of only a selected output-side interval (5 of 16 layers) carries the same reasoning accuracy as full-decoder recurrence, at a marginal cost of 5 instead of 16 decoder-layer applications per refinement step. Training uses a progressive curriculum that removes visible chain-of-thought tokens one at a time and adds latent refinements, culminating in answer-only generation. On Deep ListOps, ProsQA, and PrOntoQA, Penelope matches or nearly matches the full-decoder latent baseline (within 0.54, 1.6

Load-bearing premise

The fixed 8-slot latent memory (with 2 readout registers) is assumed to be expressive enough to hold and refine all problem-state information the tasks need; if a task's state cannot be compressed into 8 vectors, the localized loop will saturate and accuracy will fall.

Editorial extensions

If this is right

  • Additional latent reasoning depth can be bought at r/L of the full-decoder cost, making test-time compute scaling cheaper on existing decoder-only models.
  • A small fixed latent interface (8 slots) is sufficient to carry problem state for structured reasoning, suggesting memory anchors can act as a persistent reasoning scratchpad.
  • Visible chain-of-thought can be progressively distilled into the latent loop with negligible accuracy loss on these tasks, so long visible traces are not strictly necessary.
  • Localization is robust to where the recurrent interval is placed (early/middle/output-side within 0.19 points), so the method is not tied to a particular layer choice.
  • The main empirical win is a large latency reduction at comparable accuracy, not a new accuracy ceiling; additional steps are not monotonic and validation-selected depths are best.

Reading between the lines

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

  • The fixed m=8 interface is the most likely bottleneck: if a task requires more state than 8 anchors can encode, accuracy would saturate; an obvious extension is to measure accuracy vs m on tasks with larger state requirements.
  • The efficiency advantage depends on the one-time lower-prefix cost being amortized; for very long prompts or very small K, the relative benefit may shrink, a regime not analyzed in the paper.
  • Because the loop's marginal cost is only r layers, the method pairs naturally with adaptive-depth halting mechanisms, which the paper leaves open.
  • Since placement mattered little on Deep ListOps, an input-dependent learned interval is a plausible extension, though no evidence yet shows it would improve the tradeoff.
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

3 major / 4 minor

Summary. Penelope introduces a latent-reasoning framework for decoder-only Transformers in which the lower decoder prefix is evaluated once to form a boundary memory, and a selected output-side decoder interval is then iterated to refine a fixed-size latent interface (8 memory anchors, 2 readout anchors) before a single answer-context pass. A progressive CoT-to-latent curriculum transfers visible reasoning into this recurrent path. Experiments on Deep ListOps, ProsQA, and PrOntoQA with Llama-3.2-1B, plus a Qwen3.5-0.8B control, compare Penelope against Visible CoT, Coconut, CODI, and a full-decoder recurrence reference. The paper reports competitive exact-match accuracy with reduced measured inference latency, and includes matched source checkpoints, controlled optimizer-update/data-exposure budgets, audited splits, paired latency measurements, an explicit pre-answer serial-depth accounting formula, and ablation/placement controls.

Significance. If the reported results hold, Penelope offers a practically useful way to allocate test-time compute in latent space without repeatedly executing the full decoder, which is a meaningful step beyond full-decoder latent-recurrence methods. The paper's strengths include the careful matched-source evaluation, the synchronized latency measurements, the explicit serial-depth accounting in Eq. (15), and the mechanism controls (K=0 diagnostic, transition-unit ablation, and placement analysis in Table 2). The central caveat is that the headline efficiency gain is computed under an asymmetric depth-selection protocol; the paper's own symmetric comparison in Figure 2 shows a much smaller, though still positive, latency reduction.

major comments (3)
  1. [Table 1(a), §4.1–4.2, Figure 2] The headline latency reduction is not based on a symmetric tuning rule. Table 1(a) compares Penelope at validation-selected K=5 with Coconut at fixed K=8 (46 vs 160 sequential decoder-layer applications; 99.82 vs 188.15 ms). Figure 2, however, applies the same smallest-best validation rule to both recurrent paths and reports Coconut's validation optimum at K=3 (55.50% EM, 80 layer applications, 123.28 ms). At the matched operating points the latency reduction is 105.10 vs 123.28 ms (14.7%), not 46.9%, and the workload reduction is 42.5% rather than 71%. Please report three-run test EM and latency for Coconut at K=3, and either use that point in the primary table or explicitly justify why Coconut's K is fixed at 8 while Penelope's is validation-selected.
  2. [§4.1, Table 1(b,c)] The same asymmetry appears on ProsQA and PrOntoQA: Coconut uses 6 latent steps (its native schedule), while Penelope selects the smallest validation-optimal depth from K∈{1,...,6}. No validation sweep or selection curve is reported for Coconut on these tasks. If the intended protocol is a matched comparison, Coconut's validation-optimal operating point should be reported; if the fixed values are deliberate choices from the original methods, that should be stated explicitly. This is load-bearing for the claimed cross-task efficiency advantage.
  3. [§3.5, Eq. (2)] The paper never varies the latent interface size (m=8 memory slots, q=2 readout anchors) nor measures how much information the boundary memory actually carries. The claim that 'latent refinement can be localized' assumes that the fixed-size interface is sufficient for the problem state. The placement control in Table 2 shows localization is sufficient at this capacity, but a capacity sweep (varying m and q) would be needed to support the broader claim that localized refinement remains accuracy-preserving beyond the specific benchmarks. Without such evidence, the generality claim should be tempered.
minor comments (4)
  1. [§4.2] The sentence 'The principal advantage... a 46.9% reduction' conflicts with the symmetric comparison reported in Figure 2. Please clarify that this is a paired comparison against a non-validation-optimal Coconut configuration, and present the symmetric result as the primary evidence for the efficiency claim.
  2. [Table 1 caption] Add a footnote explaining why Coconut's K values in Table 1 are not validation-selected, and cross-reference Figure 2 for the validation-matched comparison.
  3. [Figure 2] Please report test (not just validation) EM for Coconut at K=3 and for Penelope at its selected K, and state whether the K=3 point is included in the three-run replication in Table 3(b).
  4. [§3.2, Eq. (15)] The accounting formula C_preans_Penelope(K)=L+(K+1)r is correct for the evaluated topology, but the derivation is terse. A one-line derivation using ℓ_s = L−r would help readers verify the count.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: efficiency accounting is definitional and external benchmark evidence is self-contained.

full rationale

We examined the derivation chain for self-definitional, fitted-input-called-prediction, self-citation, ansatz-smuggling, and renaming patterns. Equations 15–16 are an accounting definition of sequential decoder-layer applications (C_preans = L + (K+1)r) and are explicitly labeled a serial-depth proxy; they are not an empirical prediction, and the marginal-cost claim is a property of the architecture by definition, not a circular derivation. The latent-interface equations (2)–(12) define the proposed architecture, and the evaluation uses external benchmarks (Deep ListOps, ProsQA, PrOntoQA) and independent baselines (Coconut, CODI, Visible CoT) under matched source checkpoints and update budgets. Validation-selected depths are disclosed before test evaluation; the paper also reports a symmetric 'smallest-best' budget sweep in Figure 2, which yields different operating points. The asymmetry between Penelope's validation-selected K and Coconut's fixed K is a benchmarking/fairness concern, not a constructional circularity, because selecting K does not force the test accuracy or latency through an equation. The K=0 diagnostic and placement control compare the same architecture at different settings and are honestly labeled as not training-matched. No load-bearing self-citations or imported uniqueness theorems appear; cited related work is not by the present authors. We find no step where a claimed result is equivalent to its input by construction.

Assumptions & free parameters 6 free parameters · 4 assumptions · 3 invented entities

The method's efficiency accounting (Eq. 15) is arithmetic and exact given L and r, but the accuracy guarantee rests on a set of hand-chosen architectural capacities (m=8, q=2, interval [11,16)) and a curriculum schedule that are not derived from first principles and are only partially ablated. The external benchmarks are standard for the subfield, which keeps the circularity burden low.

free parameters (6)
  • Latent memory slots m = 8
    Fixed number of memory anchors A in Eq. 2; chosen by hand in §3.5; controls capacity of the boundary interface. Never ablated.
  • Readout anchors q = 2
    Fixed number of readout anchors Q in Eq. 2; chosen by hand; controls how much state the answer path sees.
  • Recurrent interval [l_s, l_e) = [11,16) (r=5)
    Selected decoder interval in §3.5; placement evaluated in Table 2 but a single interval is used for the main results; the optimum is stated to be backbone-dependent.
  • Validation-selected depth K per run = Deep ListOps (5,4,2); ProsQA (1,1,6); PrOntoQA (3,3,1)
    K is selected per run as the smallest validation-optimal depth (§4.1). This is a data-driven choice, disclosed, but not a fixed constant.
  • Time-modulation/gate constants = 0.1, 0.5, 0.5
    Coefficients in Eqs. 6, 8, and 10 (0.1 tanh, 0.5 tanh, 0.5 alpha) are fixed by hand; they control update magnitude.
  • Curriculum schedule stages = K = u = 1..6 equal blocks, then K=6
    Staging chosen in §3.4; no sensitivity analysis beyond a supplement reference.
assumptions (4)
  • domain assumption Pretrained checkpoints Llama-3.2-1B and Qwen3.5-0.8B are used as given; their internal representations are adequate for structured reasoning after LoRA and latent training.
    Introduced in §4.1; the entire method builds on these backbones; failure of the backbone would invalidate the comparison.
  • domain assumption The three benchmark tasks (Deep ListOps, ProsQA, PrOntoQA) measure the structured reasoning abilities the paper claims to preserve.
    §4.1; if these synthetic tasks are not representative, the accuracy-efficiency tradeoff may not generalize.
  • ad hoc to paper Cross-entropy loss on the curriculum target (Eq. 13-14) trains the latent path sufficiently; no auxiliary loss on latent states is needed.
    §3.4; the training objective is designed for this method and is not independently validated.
  • standard math KV caching (Cx in Eq. 3) of the lower-prefix states exactly reproduces the computation that would occur if the full decoder processed the prompt at every step.
    Assumed in §3.2; the method relies on cache correctness; no numerical verification is shown in the text.
invented entities (3)
  • Persistent boundary memory (M_t, 8 slots)
    purpose: Stores problem-conditioned state between recurrent refinement steps (Eqs. 2, 9-10).
    Only evidence is the paper's benchmark results; no separate test isolates this entity; capacity is not varied.
  • Recurrent readout state (R_t, 2 registers)
    purpose: Maintains an answer-accessible summary of the trajectory (Eqs. 11-12).
    Internal component with no external falsifiable handle beyond the paper's own evaluations.
  • Time network g_tau and residual adapter A_phi
    purpose: Provide step-dependent modulation and residual updates (Eqs. 5-7).
    Motivated by the need to distinguish recurrent steps; evidence is the aggregate ablation in §4.3, not an external prediction.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Penelope: Localized Latent Recurrence for Efficient Structured Reasoning." pith.science (2026). https://pith.science/paper/N7G2QCY5

@misc{pith2026260725915,
  author       = {Pith},
  title        = {Pith review of: Penelope: Localized Latent Recurrence for Efficient Structured Reasoning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N7G2QCY5}},
  note         = {Machine review of arXiv:2607.25915}
}
read the original abstract

Complex structured reasoning tasks often require additional computation, yet current language models obtain it mainly by increasing parameter scale or by serializing intermediate steps as chain-of-thought (CoT) tokens. The former raises training and deployment costs, while the latter ties reasoning computation to autoregressive output length. We introduce Penelope, an efficient latent-reasoning framework for pretrained decoder-only Transformers that localizes recurrent computation to a selected decoder interval. The lower decoder prefix is evaluated once to construct a problem-conditioned boundary memory, which is then iteratively refined through time-modulated GRU dynamics and recurrent readout states before answer generation. A progressive CoT-to-latent curriculum transfers visible reasoning into this internal recurrent path, allowing additional computation to be allocated in latent space without repeatedly executing the complete decoder or generating a long intermediate trace. Experiments on open-source structured-reasoning benchmarks show that, at validation-selected latent budgets, Penelope attains competitive accuracy relative to established latent-reasoning models while reducing measured inference latency. These results show that latent refinement can be localized to a narrow decoder interval, reducing repeated full-decoder execution without generating a long visible reasoning trace and providing a practical accuracy-efficiency tradeoff for decoder-only Transformer models.

Figures

Figures reproduced from arXiv: 2607.25915 by the authors.

Figure 1
Figure 1. Overview of Penelope. A: a one-time lower prefix forms problem-conditioned boundary states Hx, constructs the prompt-side cache Cx, and initializes (M0, R0). The shared output-side interval refines only this interface for K steps before one answer-context pass. B: Hx remains fixed across steps, while time-modulated memory and recurrent readout updates produce (Mt, Rt). reasoning computation corresponds to more local… view at source ↗
Figure 2
Figure 2. Runtime-verified BF16 validation efficiency for one matched Penelope/Coconut checkpoint. (a) Exact match versus [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 4 canonical work pages

  1. [1]

    Advances in Neural Information Processing Systems , volume =

    Attention Is All You Need , author =. Advances in Neural Information Processing Systems , volume =. 2017 , url =

    2017

  2. [2]

    Advances in Neural Information Processing Systems , volume =

    Chain-of-Thought Prompting Elicits Reasoning in Large Language Models , author =. Advances in Neural Information Processing Systems , volume =. 2022 , url =

    2022

  3. [3]

    Advances in Neural Information Processing Systems , volume =

    Large Language Models are Zero-Shot Reasoners , author =. Advances in Neural Information Processing Systems , volume =. 2022 , url =

    2022

  4. [4]

    International Conference on Learning Representations , year =

    Self-Consistency Improves Chain of Thought Reasoning in Language Models , author =. International Conference on Learning Representations , year =

  5. [5]

    , booktitle =

    Zelikman, Eric and Wu, Yuhuai and Mu, Jesse and Goodman, Noah D. , booktitle =. 2022 , url =

    2022

  6. [6]

    2016 , eprint =

    Adaptive Computation Time for Recurrent Neural Networks , author =. 2016 , eprint =

    2016

  7. [7]

    Proceedings of the 8th ICML Workshop on Automated Machine Learning , year =

    PonderNet: Learning to Ponder , author =. Proceedings of the 8th ICML Workshop on Automated Machine Learning , year =

  8. [8]

    International Conference on Learning Representations , year =

    Universal Transformers , author =. International Conference on Learning Representations , year =

Show all 38 references
  1. [9]

    and Salakhutdinov, Ruslan , booktitle =

    Dai, Zihang and Yang, Zhilin and Yang, Yiming and Carbonell, Jaime and Le, Quoc V. and Salakhutdinov, Ruslan , booktitle =. 2019 , doi =

    2019

  2. [10]

    Advances in Neural Information Processing Systems , volume =

    Block-Recurrent Transformers , author =. Advances in Neural Information Processing Systems , volume =. 2022 , url =

    2022

  3. [11]

    Advances in Neural Information Processing Systems , volume =

    Recurrent Memory Transformer , author =. Advances in Neural Information Processing Systems , volume =. 2022 , url =

    2022

  4. [12]

    Second Conference on Language Modeling , year =

    Training Large Language Models to Reason in a Continuous Latent Space , author =. Second Conference on Language Modeling , year =

  5. [13]

    2025 , address =

    Shen, Zhenyi and Yan, Hanqi and Zhang, Linhai and Hu, Zhanghao and Du, Yali and He, Yulan , booktitle =. 2025 , address =. doi:10.18653/v1/2025.emnlp-main.36 , url =

    2025 doi

  6. [14]

    2025 , address =

    Jiang, Nan and Wu, Ziming and Zhan, De-Chuan and Lai, Fuming and Lian, Shaobing , booktitle =. 2025 , address =. doi:10.18653/v1/2025.emnlp-main.256 , url =

    2025 doi

  7. [15]

    First Conference on Language Modeling , year =

    Quiet-STaR: Language Models Can Teach Themselves to Think Before Speaking , author =. First Conference on Language Modeling , year =

  8. [16]

    Proceedings of the 39th International Conference on Machine Learning , series =

    The CLRS Algorithmic Reasoning Benchmark , author =. Proceedings of the 39th International Conference on Machine Learning , series =. 2022 , publisher =

    2022

  9. [17]

    Advances in Neural Information Processing Systems Datasets and Benchmarks Track , year =

    Minigrid and Miniworld: Modular and Customizable Reinforcement Learning Environments for Goal-Oriented Tasks , author =. Advances in Neural Information Processing Systems Datasets and Benchmarks Track , year =

  10. [18]

    Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence , pages =

    Transformers as Soft Reasoners over Language , author =. Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence , pages =. 2020 , doi =

    2020

  11. [19]

    International Conference on Learning Representations , year =

    Long Range Arena: A Benchmark for Efficient Transformers , author =. International Conference on Learning Representations , year =

  12. [20]

    Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Student Research Workshop , pages =

    ListOps: A Diagnostic Dataset for Latent Tree Learning , author =. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Student Research Workshop , pages =. 2018 , url =

    2018

  13. [21]

    International Conference on Learning Representations , year =

    Language Models Are Greedy Reasoners: A Systematic Formal Analysis of Chain-of-Thought , author =. International Conference on Learning Representations , year =

  14. [22]

    Dubey, Abhimanyu and Jauhri, Abhinav and Pandey, Abhinav and Kadian, Abhishek and Al-Dahle, Ahmad and Letman, Aiesha and Mathur, Akhil and Schelten, Alan and Yang, Amy and Fan, Angela and others , journal =. The. 2024 , url =

    2024

  15. [23]

    2026 , howpublished =

    2026

  16. [24]

    2026 , eprint =

    Bridging the Gap Between Latent and Explicit Reasoning with Looped Transformers , author =. 2026 , eprint =

    2026

  17. [25]

    2026 , eprint =

    Training-Free Looped Transformers , author =. 2026 , eprint =

    2026

  18. [26]

    Proceedings of the 38th Canadian Conference on Artificial Intelligence , year =

    Intra-Layer Recurrence in Transformers for Language Modeling , author =. Proceedings of the 38th Canadian Conference on Artificial Intelligence , year =. doi:10.21428/594757db.834c24c6 , url =

    doi

  19. [27]

    2026 , month = jul, address =

    Fu, Renyu and Luo, Guibo , booktitle =. 2026 , month = jul, address =. doi:10.18653/v1/2026.acl-long.320 , url =

    2026 doi

  20. [28]

    The Thirteenth International Conference on Learning Representations , year =

    Reasoning with Latent Thoughts: On the Power of Looped Transformers , author =. The Thirteenth International Conference on Learning Representations , year =

  21. [29]

    Advances in Neural Information Processing Systems , volume =

    Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation , author =. Advances in Neural Information Processing Systems , volume =. 2025 , url =

    2025

  22. [30]

    International Conference on Learning Representations , year =

    LoRA: Low-Rank Adaptation of Large Language Models , author =. International Conference on Learning Representations , year =

  23. [31]

    Journal of Machine Learning Research , volume =

    Scaling Instruction-Finetuned Language Models , author =. Journal of Machine Learning Research , volume =. 2024 , url =

    2024

  24. [32]

    2024 , eprint =

    TinyLlama: An Open-Source Small Language Model , author =. 2024 , eprint =

    2024

  25. [33]

    Proceedings of the 29th Symposium on Operating Systems Principles , pages =

    Efficient Memory Management for Large Language Model Serving with PagedAttention , author =. Proceedings of the 29th Symposium on Operating Systems Principles , pages =. 2023 , doi =

    2023

  26. [34]

    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

  27. [35]

    Advances in Neural Information Processing Systems , volume =

    Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach , author =. Advances in Neural Information Processing Systems , volume =. 2025 , url =

    2025

  28. [36]

    2026 , url =

    Jeddi, Ahmadreza and Ciccone, Marco and Taati, Babak , booktitle =. 2026 , url =

    2026

  29. [37]

    ICML 2026 Workshop on Adaptive Foundation Models (AdaptFM) , year =

    Fixed-Point Reasoners: Stable and Adaptive Deep Looped Transformers , author =. ICML 2026 Workshop on Adaptive Foundation Models (AdaptFM) , year =

    2026

  30. [38]

    doi:10.48550/arXiv.2606.18023 , url =

    Yang, Jian and Guo, Shawn and Zhang, Wei and Zheng, Tianyu and Du, Yaxin and Li, Haau-Sing and Wu, Jiajun and Song, Yue and Xing, Yan and Cai, Qingsong and Huang, Zelong and Hao, Chuan and Tao, Ran and Liu, Xianglong and Zhao, Wayne Xin and Tang, Mingjie and Lv, Weifeng and Zh...

    doi arXiv

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.