pith. machine review for the scientific record. sign in

arxiv: 2605.06322 · v1 · submitted 2026-05-07 · 💻 cs.LG

Recognition: unknown

SMolLM: Small Language Models Learn Small Molecular Grammar

Akhil Jindal , Harang Ju
Authors on Pith no claims yet

Pith reviewed 2026-05-08 12:54 UTC · model grok-4.3

classification 💻 cs.LG
keywords SMILES generationsmall language modelsmolecular grammarmechanistic interpretabilityattention headstransformer modelschemical validity
0
0 comments X

The pith

A 53K-parameter transformer generates valid SMILES by resolving constraints in fixed order: brackets first, rings second, valence last.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

SMolLM is a weight-shared transformer with 53 thousand parameters that generates novel SMILES strings for drug-like molecules. It reaches 95 percent validity on the ZINC-250K benchmark while outperforming a standard GPT model with ten times more parameters. The model resolves SMILES constraints iteratively across passes in a fixed sequence, matching brackets first, handling rings second, and enforcing valence last. This ordered behavior is identified consistently through error classification, linear probing of representations, and sparse autoencoder analysis. Systematic ablation across heads and passes further shows that the initial bracket-matching step is performed by a single attention head.

Core claim

The same transformer block resolves SMILES constraints across passes in a fixed order—brackets first, rings second, and valence last—with the bracket-matching step localized to a single attention head, as shown by error classification, linear probing, and sparse autoencoders, yielding a compact mechanistically interpretable molecular generator.

What carries the argument

Fixed-order iterative constraint resolution across passes within the weight-shared transformer block, with bracket matching localized to one attention head.

If this is right

  • The approach yields a compact and mechanistically interpretable molecular generator.
  • It serves as a testbed for studying iterative computation in formal-language domains.
  • Constraint resolution occurs in a consistent sequence that can be localized to specific attention heads.
  • Small models can achieve high validity on structured generation tasks by learning grammar rules explicitly.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The fixed sequential order may generalize as a strategy for transformers learning other nested formal languages such as programming syntax.
  • Targeted interventions on specific heads could further improve validity rates in molecular design applications.
  • The success with so few parameters suggests that explicit grammar learning enables parameter-efficient models for scientific structured data.

Load-bearing premise

Linear probing, sparse autoencoders, and error classification reveal the model's actual causal computations for resolving constraints rather than surface correlations, and high benchmark validity reflects genuine grammar learning.

What would settle it

Ablating the identified single attention head for bracket matching and observing no corresponding rise in bracket-related errors in generated SMILES strings.

Figures

Figures reproduced from arXiv: 2605.06322 by Akhil Jindal, Harang Ju.

Figure 1
Figure 1. Figure 1: Overview of SmolLM. Top track: for each emitted token, the same 53K shared block runs eight passes. By truncating inference at intermediate passes, we find that the shared block solves grammar in stages, with each stage retained as depth increases: brackets by pass 2, rings by pass 4, valence by pass 8. Bottom track: the model autoregressively emits T SMILES tokens. The molecule at step T is benzimidazole,… view at source ↗
Figure 2
Figure 2. Figure 2: Pareto frontier. Weight-shared models dominate unshared GPTs below 1M parameters. view at source ↗
Figure 5
Figure 5. Figure 5: WS-53K WS-206K Method Property peak pass peak pass Probing Bracket depth 98.6% 4 99.5% 2 Ring state 97.6% 6 98.0% 5 SAE Bracket detector 0.76 2 0.79 2 Ring-digit detector 0.92 1 0.85 5 Bracket depth 0.67 6 0.63 4 Atom identity 0.68 6 0.55 7 Ring state 0.67 6 0.62 7 Linear probing. We train linear probes on each pass of the 8-pass model to test when bracket depth and ring state become decodable. In both mod… view at source ↗
Figure 3
Figure 3. Figure 3: Bracket errors collapse by pass 2, rings by pass 4, valence by pass 8 (companion to Table 1). view at source ↗
Figure 4
Figure 4. Figure 4: Ablation heatmap (head × pass, change in validity in percentage points; n=2,000 per condition for both models). WS-206K: single hot cell at the bracket head, pass 1. WS-53K: heat spreads across the bracket head, passes 1–3. 16 view at source ↗
Figure 5
Figure 5. Figure 5: Representation organization across passes. Panel (a) shows probe accuracy; panel (b) shows view at source ↗
read the original abstract

Language models for molecular design have scaled to hundreds of millions of parameters, yet how they learn chemical grammar is poorly understood. We train SMolLM, a 53K-parameter weight-shared transformer, to generate novel SMILES with 95% validity on the ZINC-250K drug-like-molecule benchmark, outperforming a standard GPT with 10 times more parameters. Mechanistically, the same block resolves SMILES constraints across passes in a fixed order: brackets first, rings second, and valence last, as shown by error classification, linear probing, and sparse autoencoders. A systematic ablation across attention heads and passes further localizes the first bracket-matching step to a single attention head. Together, these results yield a compact, mechanistically interpretable molecular generator and a testbed for studying iterative computation in formal-language domains.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The paper introduces SMolLM, a 53K-parameter weight-shared transformer trained to generate novel SMILES strings with 95% validity on the ZINC-250K benchmark, outperforming a standard GPT with 10x more parameters. It claims that the model resolves SMILES constraints across passes in a fixed order—brackets first, rings second, valence last—as evidenced by error classification, linear probing, and sparse autoencoders, with systematic ablations localizing the initial bracket-matching step to a single attention head. This yields a compact, mechanistically interpretable molecular generator and testbed for iterative computation in formal languages.

Significance. If the mechanistic claims hold, the work provides a notably small high-validity model for molecular design and a useful testbed for studying how transformers acquire formal grammars through iterative passes. The small parameter count, high validity rate, and use of multiple converging interpretability methods (error classification, probing, SAEs) are strengths that could advance interpretable AI for chemistry. However, the significance for mechanistic understanding is reduced because the evidence remains correlational rather than causal.

major comments (3)
  1. [Abstract and mechanistic analysis] The central claim that the same block resolves SMILES constraints in a fixed order (brackets first, rings second, valence last) rests on error classification of generated strings. This identifies which constraint fails at output but does not establish that the model internally resolves them sequentially across passes; the observed error distribution could equally arise from training data biases or output statistics rather than ordered internal computation (Abstract and mechanistic analysis).
  2. [Mechanistic interpretability section] Linear probing and sparse autoencoders are used to detect features correlated with bracket/ring/valence states and to localize computation. While these methods recover linearly separable or sparse features, their presence does not entail that the model uses the information in the claimed sequence or that the identified head performs the matching operation (mechanistic interpretability section).
  3. [Ablation study] The ablation across attention heads and passes localizes bracket-matching to a single head in the first pass. However, performance drops upon head removal could reflect general capacity loss or downstream effects rather than specific causal localization; a selective intervention (e.g., activation patching at the bracket stage) that increases bracket errors while leaving ring/valence errors largely unchanged would be required to support the claim (ablation study).
minor comments (1)
  1. [Methods] The abstract and methods could provide more explicit details on the training schedule, loss weighting, and exact architecture (e.g., number of layers, head dimensions) to aid reproducibility, as these are listed among the free parameters.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the constructive and detailed feedback, which highlights key distinctions between correlational and causal evidence in our mechanistic claims. We address each major comment point-by-point below, providing clarifications and indicating revisions to better qualify our conclusions while preserving the paper's contributions on the compact model and interpretability testbed.

read point-by-point responses

  1. Referee: The central claim that the same block resolves SMILES constraints in a fixed order (brackets first, rings second, valence last) rests on error classification of generated strings. This identifies which constraint fails at output but does not establish that the model internally resolves them sequentially across passes; the observed error distribution could equally arise from training data biases or output statistics rather than ordered internal computation (Abstract and mechanistic analysis).

    Authors: We agree that error classification alone is correlational and could reflect output statistics or data biases. Our full analysis integrates this with linear probing (showing constraint features emerging progressively across passes) and sparse autoencoders (extracting distinct bracket/ring/valence features). The ablation provides further localization. We will revise the abstract and mechanistic analysis section to state that the results are consistent with ordered internal resolution based on converging evidence, rather than claiming definitive proof, and add a paragraph discussing alternative explanations such as training data biases. revision: partial

  2. Referee: Linear probing and sparse autoencoders are used to detect features correlated with bracket/ring/valence states and to localize computation. While these methods recover linearly separable or sparse features, their presence does not entail that the model uses the information in the claimed sequence or that the identified head performs the matching operation (mechanistic interpretability section).

    Authors: We concur that probing and SAEs yield correlational evidence and do not directly prove usage in sequence or that the head executes the operation. The sequence inference comes from the temporal ordering of feature activation across passes, with the head's role supported by ablation specificity. We will revise the mechanistic interpretability section to explicitly note the correlational limits of these methods, clarify that they provide consistent but not causal support for the sequence, and discuss how the multi-method approach strengthens the overall interpretation. revision: partial

  3. Referee: The ablation across attention heads and passes localizes bracket-matching to a single head in the first pass. However, performance drops upon head removal could reflect general capacity loss or downstream effects rather than specific causal localization; a selective intervention (e.g., activation patching at the bracket stage) that increases bracket errors while leaving ring/valence errors largely unchanged would be required to support the claim (ablation study).

    Authors: The ablation demonstrates that ablating the target head in pass 1 increases bracket errors far more than ablating other heads, with comparatively small effects on ring/valence errors, which is inconsistent with uniform capacity loss. We agree that activation patching would provide stronger causal evidence for localization. However, such interventions require substantial additional compute and are not feasible in this revision. We will revise the ablation study section to highlight the error-type specificity in more detail and add a limitations paragraph acknowledging the correlational nature while proposing activation patching as future work. revision: partial

standing simulated objections not resolved
  • Request for activation patching or other causal interventions to confirm the specific mechanistic role of the identified attention head in bracket matching.

Circularity Check

0 steps flagged

No circularity: claims rest on post-training empirical probes of a trained model, not on equations or self-citations that reduce to inputs.

full rationale

The paper trains SMolLM on ZINC-250K, then applies error classification, linear probing, SAEs, and head ablations to observe that constraints appear resolved in bracket-ring-valence order and that bracket matching localizes to one head. These are standard post-hoc analyses on a fixed trained network; none of the reported quantities (validity rates, probe accuracies, ablation deltas) are defined in terms of themselves or fitted parameters within the paper. No equations equate the claimed ordering to any internal definition, and no load-bearing self-citations or uniqueness theorems are invoked. The derivation chain is therefore self-contained empirical observation rather than tautological reduction.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard transformer training plus interpretability assumptions; the main added elements are the empirical ordering finding and the single-head localization rather than new theoretical entities.

free parameters (2)
  • 53K parameter count and architecture details
    Model size and layer/head configuration chosen to demonstrate efficiency; exact values are hyperparameters selected for the reported validity.
  • training schedule and loss weighting
    Hyperparameters tuned to achieve 95% validity on the benchmark.
axioms (2)
  • domain assumption SMILES validity can be reliably checked by syntactic rules for brackets, rings, and valence
    Invoked in the error classification and validity metric throughout the abstract.
  • ad hoc to paper Linear probes and sparse autoencoders recover the model's internal computation order
    Central to the mechanistic claim that constraints are resolved in a fixed sequence.

pith-pipeline@v0.9.0 · 5433 in / 1468 out tokens · 41333 ms · 2026-05-08T12:54:00.434191+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

84 extracted references · 35 canonical work pages · 14 internal anchors

  1. [2]

    Bagal, Viraj and Aggarwal, Rishal and Vinod, P. K. and Priyakumar, U. Deva , title =. Journal of Chemical Information and Modeling , volume =. 2022 , doi =

    2022

  2. [4]

    A Mechanistic Analysis of Looped Reasoning Language Models , journal =

    Blayney, Henry and Arroyo,. A Mechanistic Analysis of Looped Reasoning Language Models , journal =. 2026 , url =

    2026

  3. [5]

    and Hume, Tristan and Carter, Shan and Henighan, Tom and Olah, Christopher , title =

    Bricken, Trenton and Templeton, Adly and Batson, Joshua and Chen, Brian and Jermyn, Adam and Conerly, Tom and Turner, Nick and Anil, Cem and Denison, Carson and Askell, Amanda and Lasenby, Robert and Wu, Yifan and Kravec, Shauna and Schiefer, Nicholas and Maxwell, Tim and Joseph, Nicholas and Hatfield-Dodds, Zac and Tamkin, Alex and Nguyen, Karina and McL...

  4. [7]

    What you can cram into a single \ &!\#* vector: Probing sentence embeddings for linguistic properties , booktitle =

    Conneau, Alexis and Kruszewski, Germ. What you can cram into a single \ &!\#* vector: Probing sentence embeddings for linguistic properties , booktitle =. 2018 , abstract =

    2018

  5. [8]

    ICLR , year =

    Cunningham, Hoagy and Ewart, Aidan and Riggs, Logan and Huben, Robert and Sharkey, Lee , title =. ICLR , year =

  6. [9]

    ICLR , year =

    Dehghani, Mostafa and Gouws, Stephan and Vinyals, Oriol and Uszkoreit, Jakob and Kaiser, Lukasz , title =. ICLR , year =

  7. [10]

    Neural Networks and the

    Del. Neural Networks and the. ICLR , year =

  8. [12]

    Transformer Circuits Thread , year =

    Elhage, Nelson and Nanda, Neel and Olsson, Catherine and Henighan, Tom and Joseph, Nicholas and Mann, Ben and Askell, Amanda and Bai, Yuntao and Chen, Anna and Conerly, Tom and DasSarma, Nova and Drain, Dawn and Ganguli, Deep and Hatfield-Dodds, Zac and Hernandez, Danny and Jones, Andy and Kernion, Jackson and Lovitt, Liane and Ndousse, Kamal and Amodei, ...

  9. [13]

    Transformer Circuits Thread , year =

    Elhage, Nelson and Hume, Tristan and Olsson, Catherine and Schiefer, Nicholas and Henighan, Tom and Kravec, Shauna and Hatfield-Dodds, Zac and Lasenby, Robert and Drain, Dawn and Chen, Carol and Grosse, Roger and McCandlish, Sam and Kaplan, Jared and Amodei, Dario and Wattenberg, Martin and Olah, Christopher , title =. Transformer Circuits Thread , year =

  10. [15]

    and Papailiopoulos, Dimitris , title =

    Giannou, Angeliki and Rajput, Shashank and Sohn, Jy-yong and Lee, Kangwook and Lee, Jason D. and Papailiopoulos, Dimitris , title =. ICML , year =

  11. [16]

    ICLR , year =

    Gu, Yuxian and Dong, Li and Wei, Furu and Huang, Minlie , title =. ICLR , year =

  12. [17]

    Transactions of the Association for Computational Linguistics , volume =

    Hahn, Michael , title =. Transactions of the Association for Computational Linguistics , volume =. 2020 , abstract =

    2020

  13. [21]

    AAAI , year =

    Huang, Victor Shea-Jay and Zhuo, Le and Xin, Yi and Wang, Zhaokai and Wang, Fu-Yun and Wang, Yue and Zhang, Renrui and Gao, Peng and Li, Hongsheng , title =. AAAI , year =

  14. [22]

    and Sterling, Teague and Mysinger, Michael M

    Irwin, John J. and Sterling, Teague and Mysinger, Michael M. and Bolstad, Erin S. and Coleman, Ryan G. , title =. Journal of Chemical Information and Modeling , volume =. 2012 , doi =

    2012

  15. [25]

    Self-Referencing Embedded Strings (

    Krenn, Mario and H. Self-Referencing Embedded Strings (. Machine Learning: Science and Technology , volume =. 2020 , abstract =

    2020

  16. [27]

    ICLR , year =

    Lan, Zhenzhong and Chen, Mingda and Goodman, Sebastian and Gimpel, Kevin and Sharma, Piyush and Soricut, Radu , title =. ICLR , year =

  17. [29]

    ICML , year =

    Liu, Zechun and Zhao, Changsheng and Iandola, Forrest and Lai, Chen and Tian, Yuandong and Fedorov, Igor and Xiong, Yunyang and Chang, Ernie and Shi, Yangyang and Krishnamoorthi, Raghuraman and Chandra, Vikas , title =. ICML , year =

  18. [31]

    COLING , year =

    Niu, Jingcheng and Lu, Wenjie and Penn, Gerald , title =. COLING , year =

  19. [32]

    Journal of Cheminformatics , volume =

    Olivecrona, Marcus and Blaschke, Thomas and Engkvist, Ola and Chen, Hongming , title =. Journal of Cheminformatics , volume =. 2017 , abstract =

    2017

  20. [33]

    Transformer Circuits Thread , year =

    Olsson, Catherine and Elhage, Nelson and Nanda, Neel and Joseph, Nicholas and DasSarma, Nova and Henighan, Tom and Mann, Ben and Askell, Amanda and Bai, Yuntao and Chen, Anna and Conerly, Tom and Drain, Dawn and Ganguli, Deep and Hatfield-Dodds, Zac and Hernandez, Danny and Johnston, Scott and Jones, Andy and Kernion, Jackson and Lovitt, Liane and Ndousse...

  21. [34]

    and Petty, Jackson and Shi, Chuan and Merrill, William and Linzen, Tal , title =

    Hu, Michael Y. and Petty, Jackson and Shi, Chuan and Merrill, William and Linzen, Tal , title =. ACL , year =

  22. [36]

    Journal of Chemical Information and Modeling , volume =

    Preuer, Kristina and Renz, Philipp and Unterthiner, Thomas and Hochreiter, Sepp and Klambauer, G. Journal of Chemical Information and Modeling , volume =. 2018 , abstract =

    2018

  23. [37]

    OpenAI Blog , year =

    Radford, Alec and Wu, Jeffrey and Child, Rewon and Luan, David and Amodei, Dario and Sutskever, Ilya , title =. OpenAI Blog , year =

  24. [38]

    and Finn, Chelsea , title =

    Rafailov, Rafael and Sharma, Archit and Mitchell, Eric and Ermon, Stefano and Manning, Christopher D. and Finn, Chelsea , title =. NeurIPS , year =

  25. [39]

    Findings of EMNLP , year =

    Reid, Machel and Marrese-Taylor, Edison and Matsuo, Yutaka , title =. Findings of EMNLP , year =

  26. [41]

    Segler, Marwin H. S. and Kogej, Thierry and Tyrchan, Christian and Waller, Mark P. , title =. ACS Central Science , volume =. 2018 , abstract =

    2018

  27. [43]

    Neurocomputing , volume =

    Su, Jianlin and Ahmed, Murtadha and Lu, Yu and Pan, Shengfeng and Bo, Wen and Liu, Yunfeng , title =. Neurocomputing , volume =. 2024 , abstract =

    2024

  28. [44]

    ACL , year =

    Tenney, Ian and Das, Dipanjan and Pavlick, Ellie , title =. ACL , year =

  29. [45]

    Circuits, Features, and Heuristics in Molecular Transformers , journal =

    Varadi, Krist. Circuits, Features, and Heuristics in Molecular Transformers , journal =. 2025 , url =

    2025

  30. [46]

    NeurIPS , year =

    Vig, Jesse and Gehrmann, Sebastian and Belinkov, Yonatan and Qian, Sharon and Nevo, Daniel and Singer, Yaron and Shieber, Stuart , title =. NeurIPS , year =

  31. [47]

    Transformers Learn In-Context by Gradient Descent , booktitle =

    von Oswald, Johannes and Niklasson, Eyvind and Randazzo, Ettore and Sacramento, Jo. Transformers Learn In-Context by Gradient Descent , booktitle =. 2023 , url =

    2023

  32. [48]

    NeurIPS , year =

    Wen, Kaiyue and Li, Yuchen and Liu, Bingbin and Risteski, Andrej , title =. NeurIPS , year =

  33. [50]

    Chemical Transformer Compression for Accelerating Both Training and Inference of Molecular Modeling , journal =

    Yu, Yi and B. Chemical Transformer Compression for Accelerating Both Training and Inference of Molecular Modeling , journal =. 2022 , url =

    2022

  34. [54]

    Relaxed recursive transformers: Effective parameter sharing with layer-wise lora.arXiv preprint arXiv:2410.20672, 2024

    Sangmin Bae, Adam Fisch, Hrayr Harutyunyan, Ziwei Ji, Seungyeon Kim, and Tal Schuster. Relaxed recursive transformers: Effective parameter sharing with layer-wise LoRA . arXiv preprint arXiv:2410.20672, 2025. URL https://arxiv.org/abs/2410.20672

    work page arXiv 2025

  35. [55]

    Viraj Bagal, Rishal Aggarwal, P. K. Vinod, and U. Deva Priyakumar. MolGPT : Molecular generation using a transformer-decoder model. Journal of Chemical Information and Modeling, 62 0 (9): 0 2064--2076, 2022. doi:10.1021/acs.jcim.1c00600

    work page doi:10.1021/acs.jcim.1c00600 2064

  36. [56]

    Smiles enumeration as data augmentation for neural network modeling of molecules.ArXiv, abs/1703.07076, 2017

    Esben Jannik Bjerrum. SMILES enumeration as data augmentation for neural network modeling of molecules. arXiv preprint arXiv:1703.07076, 2017. URL https://arxiv.org/abs/1703.07076

    work page arXiv 2017

  37. [57]

    A Mechanistic Analysis of Looped Reasoning Language Models

    Henry Blayney, \'A lvaro Arroyo, Johan Obando-Ceron, Pablo Samuel Castro, Aaron Courville, Michael M. Bronstein, and Xiaowen Dong. A mechanistic analysis of looped reasoning language models. arXiv preprint arXiv:2604.11791, 2026. URL https://arxiv.org/abs/2604.11791

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  38. [58]

    Burke, Tristan Hume, Shan Carter, Tom Henighan, and Christopher Olah

    Trenton Bricken, Adly Templeton, Joshua Batson, Brian Chen, Adam Jermyn, Tom Conerly, Nick Turner, Cem Anil, Carson Denison, Amanda Askell, Robert Lasenby, Yifan Wu, Shauna Kravec, Nicholas Schiefer, Tim Maxwell, Nicholas Joseph, Zac Hatfield-Dodds, Alex Tamkin, Karina Nguyen, Brayden McLean, Josiah E. Burke, Tristan Hume, Shan Carter, Tom Henighan, and C...

    2023

  39. [59]

    Hasson, and S

    Jonathan Cohen, Avi G. Hasson, and S. Tanovic. Unveiling latent knowledge in chemistry language models through sparse autoencoders. arXiv preprint arXiv:2512.08077, 2025. URL https://arxiv.org/abs/2512.08077

    work page arXiv 2025

  40. [60]

    What you can cram into a single \ &!\#* vector: Probing sentence embeddings for linguistic properties

    Alexis Conneau, Germ \'a n Kruszewski, Guillaume Lample, Lo \"i c Barrault, and Marco Baroni. What you can cram into a single \ &!\#* vector: Probing sentence embeddings for linguistic properties. In Proceedings of ACL, 2018

    2018

  41. [61]

    Sparse Autoencoders Find Highly Interpretable Features in Language Models

    Hoagy Cunningham, Aidan Ewart, Logan Riggs, Robert Huben, and Lee Sharkey. Sparse autoencoders find highly interpretable features in language models. In ICLR, 2024. URL https://arxiv.org/abs/2309.08600

    work page internal anchor Pith review arXiv 2024

  42. [62]

    Universal Transformers

    Mostafa Dehghani, Stephan Gouws, Oriol Vinyals, Jakob Uszkoreit, and Lukasz Kaiser. Universal transformers. In ICLR, 2019. URL https://arxiv.org/abs/1807.03819

    work page internal anchor Pith review arXiv 2019

  43. [63]

    Gr \'e goire Del \'e tang, Anian Ruoss, Jordi Grau-Moya, Tim Genewein, Li Kevin Wenliang, Elliot Catt, Chris Cundy, Marcus Hutter, Shane Legg, Joel Veness, and Pedro A. Ortega. Neural networks and the Chomsky hierarchy. In ICLR, 2023

    2023

  44. [64]

    Are Latent Reasoning Models Easily Interpretable?

    Nicholas Dilgren and Sarah Wiegreffe. Are latent reasoning models easily interpretable? T race-decoding analysis of looped transformers. arXiv preprint arXiv:2604.04902, 2026. URL https://arxiv.org/abs/2604.04902

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  45. [65]

    A mathematical framework for transformer circuits

    Nelson Elhage, Neel Nanda, Catherine Olsson, Tom Henighan, Nicholas Joseph, Ben Mann, Amanda Askell, Yuntao Bai, Anna Chen, Tom Conerly, Nova DasSarma, Dawn Drain, Deep Ganguli, Zac Hatfield-Dodds, Danny Hernandez, Andy Jones, Jackson Kernion, Liane Lovitt, Kamal Ndousse, Dario Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, and Christopher O...

    2021

  46. [66]

    Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach

    Jonas Geiping, Sean McLeish, Neel Jain, John Kirchenbauer, Siddharth Singh, Brian R. Bartoldson, Bhavya Kailkhura, Abhinav Bhatele, and Tom Goldstein. Scaling up test-time compute with latent reasoning: A recurrent depth approach. arXiv preprint arXiv:2502.05171, 2025. URL https://arxiv.org/abs/2502.05171

    work page internal anchor Pith review arXiv 2025

  47. [67]

    4) Giannou, A., Rajput, S., Sohn, J.-y., Lee, K., Lee, J

    Angeliki Giannou, Shashank Rajput, Jy-yong Sohn, Kangwook Lee, Jason D. Lee, and Dimitris Papailiopoulos. Looped transformers as programmable computers. In ICML, 2023. URL https://arxiv.org/abs/2301.13196

    work page arXiv 2023

  48. [68]

    ELT: Elastic Looped Transformers for Visual Generation

    Anand Goyal, Aishwarya Agrawal, Cem Anil, Tanvi Jain, Aishwarya Paul, and Aditya Kusupati. ELT : Elastic looped transformers for visual generation. arXiv preprint arXiv:2604.09168, 2026. URL https://arxiv.org/abs/2604.09168

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  49. [69]

    Theoretical limitations of self-attention in neural sequence models

    Michael Hahn. Theoretical limitations of self-attention in neural sequence models. Transactions of the Association for Computational Linguistics, 8: 0 156--171, 2020

    2020

  50. [70]

    Sparse autoencoders can interpret randomly initialized transformers

    Thomas Heap, Tim Lawson, Lucy Farnik, and Laurence Aitchison. Sparse autoencoders can interpret randomly initialized transformers. arXiv preprint arXiv:2501.17727, 2025. URL https://arxiv.org/abs/2501.17727

    work page arXiv 2025

  51. [71]

    Distilling the Knowledge in a Neural Network

    Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531, 2015. NeurIPS 2014 Deep Learning Workshop

    work page internal anchor Pith review arXiv 2015

  52. [72]

    Hu, Jackson Petty, Chuan Shi, William Merrill, and Tal Linzen

    Michael Y. Hu, Jackson Petty, Chuan Shi, William Merrill, and Tal Linzen. Between circuits and Chomsky : Pre-pretraining on formal languages imparts linguistic biases. In ACL, 2025. URL https://arxiv.org/abs/2502.19249

    work page arXiv 2025

  53. [73]

    TIDE : Temporal-aware sparse autoencoders for interpretable diffusion transformers in image generation

    Victor Shea-Jay Huang, Le Zhuo, Yi Xin, Zhaokai Wang, Fu-Yun Wang, Yue Wang, Renrui Zhang, Peng Gao, and Hongsheng Li. TIDE : Temporal-aware sparse autoencoders for interpretable diffusion transformers in image generation. In AAAI, 2026. URL https://arxiv.org/abs/2503.07050

    work page arXiv 2026

  54. [74]

    Irwin, Teague Sterling, Michael M

    John J. Irwin, Teague Sterling, Michael M. Mysinger, Erin S. Bolstad, and Ryan G. Coleman. ZINC : A free tool to discover chemistry for biology. Journal of Chemical Information and Modeling, 52 0 (7): 0 1757--1768, 2012. doi:10.1021/ci3001277

    work page doi:10.1021/ci3001277 2012

  55. [75]

    Gomez, Martin Menten, and Eleni Triantafillou

    Georgios Kaissis, Hendrik Mildenberger, Aidan N. Gomez, Martin Menten, and Eleni Triantafillou. Step-resolved data attribution for recurrent-depth transformers. arXiv preprint arXiv:2602.10097, 2026. URL https://arxiv.org/abs/2602.10097

    work page arXiv 2026

  56. [76]

    Loop, Think, & Generalize: Implicit Reasoning in Recurrent-Depth Transformers

    Aditya Kohli, V. Parthasarathy, Y. Sun, and S. Yao. Loop, think, and generalize: Implicit reasoning in recurrent-depth transformers. arXiv preprint arXiv:2604.07822, 2026. URL https://arxiv.org/abs/2604.07822

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  57. [77]

    Self-referencing embedded strings ( SELFIES ): A 100\ molecular string representation

    Mario Krenn, Florian H \"a se, AkshatKumar Nigam, Pascal Friederich, and Al \'a n Aspuru-Guzik. Self-referencing embedded strings ( SELFIES ): A 100\ molecular string representation. Machine Learning: Science and Technology, 1 0 (4): 0 045024, 2020

    2020

  58. [78]

    Energy-entropy regularization: The true power of minimal looped transformers

    Wai-Lun Lam. Energy-entropy regularization: The true power of minimal looped transformers. arXiv preprint arXiv:2601.09588, 2026. URL https://arxiv.org/abs/2601.09588

    work page arXiv 2026

  59. [79]

    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, 2020. URL https://arxiv.org/abs/1909.11942

    work page internal anchor Pith review arXiv 2020

  60. [80]

    Y. Li, J. Bai, S. Chen, J. Li, Z. Yang, Y. Lin, and M. Zhang. Dynamics within latent chain-of-thought: Staged functionality and non-local routing in looped reasoning. arXiv preprint arXiv:2602.08783, 2026. URL https://arxiv.org/abs/2602.08783

    work page arXiv 2026

  61. [81]

    MobileLLM: Optimizing sub-billion parameter language models for on-device use cases.arXiv preprint arXiv:2402.14905, 2024

    Zechun Liu, Changsheng Zhao, Forrest Iandola, Chen Lai, Yuandong Tian, Igor Fedorov, Yunyang Xiong, Ernie Chang, Yangyang Shi, Raghuraman Krishnamoorthi, and Vikas Chandra. MobileLLM : Optimizing sub-billion parameter language models for on-device use cases. In ICML, 2024. URL https://arxiv.org/abs/2402.14905

    work page arXiv 2024

  62. [82]

    arXiv preprint arXiv:2507.02199 , year=

    Wenquan Lu, Aakriti Kar, Maximilian Bauer, and Cem Anil. Latent chain-of-thought? D ecoding the depth-recurrent transformer. arXiv preprint arXiv:2507.02199, 2025. URL https://arxiv.org/abs/2507.02199

    work page arXiv 2025

  63. [83]

    Does BERT rediscover a classical NLP pipeline? In COLING, 2022

    Jingcheng Niu, Wenjie Lu, and Gerald Penn. Does BERT rediscover a classical NLP pipeline? In COLING, 2022

    2022

  64. [84]

    Molecular de-novo design through deep reinforcement learning

    Marcus Olivecrona, Thomas Blaschke, Ola Engkvist, and Hongming Chen. Molecular de-novo design through deep reinforcement learning. Journal of Cheminformatics, 9 0 (1): 0 48, 2017

    2017

  65. [85]

    In-context learning and induction heads

    Catherine Olsson, Nelson Elhage, Neel Nanda, Nicholas Joseph, Nova DasSarma, Tom Henighan, Ben Mann, Amanda Askell, Yuntao Bai, Anna Chen, Tom Conerly, Dawn Drain, Deep Ganguli, Zac Hatfield-Dodds, Danny Hernandez, Scott Johnston, Andy Jones, Jackson Kernion, Liane Lovitt, Kamal Ndousse, Dario Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, a...

    2022

  66. [86]

    Hayden Prairie, Zachary Novack, Taylor Berg-Kirkpatrick, and Daniel Y. Fu. Parcae: Scaling laws for stable looped language models. arXiv preprint arXiv:2604.12946, 2026. URL https://arxiv.org/abs/2604.12946

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  67. [87]

    Fr \'e chet ChemNet Distance : A metric for generative models for molecules in drug discovery

    Kristina Preuer, Philipp Renz, Thomas Unterthiner, Sepp Hochreiter, and G \"u nter Klambauer. Fr \'e chet ChemNet Distance : A metric for generative models for molecules in drug discovery. Journal of Chemical Information and Modeling, 58 0 (9): 0 1736--1741, 2018

    2018

  68. [88]

    Language models are unsupervised multitask learners

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language models are unsupervised multitask learners. OpenAI Blog, 2019

    2019

  69. [89]

    Manning, and Chelsea Finn

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D. Manning, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. In NeurIPS, 2023

    2023

  70. [90]

    Subformer: Exploring weight sharing for parameter efficiency in generative transformers,

    Machel Reid, Edison Marrese-Taylor, and Yutaka Matsuo. Subformer: Exploring weight sharing for parameter efficiency in generative transformers. In Findings of EMNLP, 2021. URL https://arxiv.org/abs/2101.00234

    work page arXiv 2021

  71. [91]

    GP-MoLFormer : A foundation model for molecular generation

    Jerret Ross, Brian Belgodere, Samuel Hoffman, Vijil Chenthamarakshan, Youssef Mroueh, and Payel Das. GP-MoLFormer : A foundation model for molecular generation. arXiv preprint arXiv:2405.04912, 2024. URL https://arxiv.org/abs/2405.04912

    work page arXiv 2024

  72. [92]

    Marwin H. S. Segler, Thierry Kogej, Christian Tyrchan, and Mark P. Waller. Generating focused molecule libraries for drug discovery with recurrent neural networks. ACS Central Science, 4 0 (1): 0 120--131, 2018

    2018

  73. [93]

    GLU Variants Improve Transformer

    Noam Shazeer. GLU variants improve transformer. arXiv preprint arXiv:2002.05202, 2020

    work page internal anchor Pith review arXiv 2002

  74. [94]

    RoFormer : Enhanced transformer with rotary position embedding

    Jianlin Su, Murtadha Ahmed, Yu Lu, Shengfeng Pan, Wen Bo, and Yunfeng Liu. RoFormer : Enhanced transformer with rotary position embedding. Neurocomputing, 568: 0 127063, 2024

    2024

  75. [95]

    BERT rediscovers the classical NLP pipeline

    Ian Tenney, Dipanjan Das, and Ellie Pavlick. BERT rediscovers the classical NLP pipeline. In ACL, 2019

    2019

  76. [96]

    Circuits, features, and heuristics in molecular transformers.arXiv [cs.LG], 2025

    Krist \'o f Varadi, M \'a t \'e Marosi, and P \'e ter Antal. Circuits, features, and heuristics in molecular transformers. arXiv preprint arXiv:2512.09757, 2025. URL https://arxiv.org/abs/2512.09757

    work page arXiv 2025

  77. [97]

    Investigating gender bias in language models using causal mediation analysis

    Jesse Vig, Sebastian Gehrmann, Yonatan Belinkov, Sharon Qian, Daniel Nevo, Yaron Singer, and Stuart Shieber. Investigating gender bias in language models using causal mediation analysis. In NeurIPS, 2020

    2020

  78. [98]

    arXiv preprint arXiv:2212.07677 , title =

    Johannes von Oswald, Eyvind Niklasson, Ettore Randazzo, Jo \ a o Sacramento, Alexander Mordvintsev, Andrey Zhmoginov, and Max Vladymyrov. Transformers learn in-context by gradient descent. In ICML, 2023. URL https://arxiv.org/abs/2212.07677

    work page arXiv 2023

  79. [99]

    Transformers are uninterpretable with myopic methods: A case study with bounded Dyck grammars

    Kaiyue Wen, Yuchen Li, Bingbin Liu, and Andrej Risteski. Transformers are uninterpretable with myopic methods: A case study with bounded Dyck grammars. In NeurIPS, 2023. URL https://arxiv.org/abs/2312.01429

    work page arXiv 2023

  80. [100]

    Y. Xu, S. Pan, Z. Yuan, Y. Li, X. Zhu, and S. Ji. Unveiling scaling behaviors in molecular language models: Effects of model size, data, and representation. arXiv preprint arXiv:2601.22757, 2026. URL https://arxiv.org/abs/2601.22757

    work page arXiv 2026

Showing first 80 references.