Stockmark-Nemotron-3-Nano-Omni-JapanDocReader: Structured Document Parsing via Capability Injection and Forgetting Control

Hayato Aida; Kosuke Arima; Makoto Morinaga; Shi Chen; Shohei Tanaka

REVIEW 4 major objections 5 minor 16 references

Stockmark-Nemotron-3-Nano-Omni-JapanDocReader: Structured Document Parsing via Capability Injection and Forgetting Control

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A two-stage recipe lifts Japanese document parsing to 87.67, without destroying document VQA.

desk verdict A credible post-training recipe for Japanese structured document parsing, but the headline scores all come from the authors' own synthetic benchmark whose metric is built from the same scoring functions as the RL reward; treat 87.67 as in-distribution until a real-document or distribution-shift test appears. read the letter →

arxiv 2608.06758 v1 pith:TL7CGPZQ submitted 2026-08-07 cs.CL

Shi Chen , Hayato Aida , Makoto Morinaga , Shohei Tanaka , Kosuke Arima This is my paper · ORCID

classification cs.CL

keywords structureddocumentparsingJapaneseunderstandingcapabilityinjectionforgettingcontrolDAPOreinforcementlearningsyntheticdataengineVQAvision-languagemodel

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper reports a post-training recipe that gives a reasoning-oriented Japanese vision-language model the ability to convert a page image into a complete structured JSON representation—layout classes, bounding boxes, text, tables as HTML, formulas as LaTeX, and picture descriptions—while preserving most of its existing ability to answer questions about Japanese documents. The authors find that fine-tuning on parsing data alone improves parsing but measurably degrades document VQA; mixing VQA data back into SFT restores most of that ability with almost no parsing loss. They then show that a token-level reinforcement-learning pass (DAPO) with a task-matched, continuous reward pushes parsing further, scoring 87.67 on the OmniDocBench-JASyn benchmark, above the base model's 70.43 and above several open-weight comparison models. The cost is some additional VQA drift, which the paper treats as the central capability-injection versus forgetting-control trade-off.

What carries the argument

The load-bearing machinery is a representation-aware reward coupled to a strict validity gate. The reward canonicalizes both prediction and reference into sets of document elements (class, bbox, contents, caption), matches predicted to ground-truth elements greedily over same-class bbox IoU, then scores schema validity, text similarity (1 minus normalized edit distance), table TEDS, formula CDM or LaTeX edit similarity, bbox IoU, and picture presence with weights 0.15, 0.45, 0.20, 0.05, 0.10, and 0.05. Any unparsable, truncated, or severely degenerate output receives zero total reward. This matching removes index-order noise so a single insertion or deletion does not misalign all later elements. Around it sit the mixed-SFT data blend (94k VQA plus 25k parsing examples) and a variance-based prompt filter that keeps only prompts whose rollout reward variance exceeds a floor and whose mean lies in a learnable band; DAPO's token-level policy-gradient update, without KL anchoring, then has precise gradients on schema, tables, and formulas.

What would settle it

Evaluate the released model on a human-annotated corpus of real scanned Japanese documents, or on any independently built Japanese parsing benchmark with different rendering and degradation statistics; if DocParse-Overall on that set does not stay clearly above the base model and the Qwen baselines, the central claim of general capability injection fails. A cheaper check: compute image- or text-level overlap between the training data engine's outputs and OmniDocBench-JASyn; near-duplicate examples would indicate evaluation leakage.

Watch

Extended reading notes

Core claim

The paper's central claim is that the capability-injection/forgetting trade-off is not fixed. Parsing-centric SFT raises DocParse-Overall from 70.43 to 86.56 while dropping VQA-Overall from 0.863 to 0.826; mixing VQA examples into the same SFT run holds parsing at 86.60 while recovering VQA to 0.844. Removing the English reasoning trace entirely (instruct mode) is catastrophic for multi-step business-document reasoning (JA-Business-Doc-RQ-Bench falls from 0.895 to 0.384), so the final recipe keeps thinking-mode traces. On top of the mixed-SFT checkpoint, DAPO parsing-centric RL improves DocParse-Overall to 87.67 and Formula CDM from 89.48 to 92.87, outperforming the other open-weight models compared; GSPO, by contrast, falls to 84.01, which the authors attribute to its sequence-level importance ratio and KL anchoring being too conservative for fine-grained parsing improvement. The released model is the mixed-SFT plus DAPO checkpoint.

Load-bearing premise

The evaluation benchmark OmniDocBench-JASyn is a synthetic Japanese document set created by the same organization and generated from the same rendering pipeline family as the training data, and the paper gives no holdout split or real-document test, so the reported superiority could reflect in-distribution fit rather than general parsing skill.

Editorial extensions

If this is right

The released model achieves the best DocParse-Overall (87.67) and best Reading Order Edit (0.2366) among the compared open-weight models, so practitioners can expect stronger tables, formulas, and reading order than the Qwen baselines, at the cost of somewhat higher pure-text edit error.
Mixed SFT preserves nearly all of the parsing gain of parsing-centric SFT while recovering VQA-Overall from 0.826 to 0.844, so mixing VQA data back into SFT is a cheap forgetting-control measure.
DAPO-based RL improves parsing beyond the SFT ceiling, which means the SFT data alone does not bound achievable structured parsing quality when a matched reward and prompt filtering are used.
Dropping reasoning traces in favor of direct JSON output is harmful for multi-step document reasoning; the paper's instruct-mode experiment implies that preserving English reasoning traces during SFT is important for retaining VQA behavior.

Reading between the lines

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

If the benchmark and training data share the same synthetic rendering pipeline, the recipe's transferability to real scanned Japanese documents is untested; a natural extension is evaluation on a human-annotated real-document parsing set.
Variance-based prompt filtering is a general principle for continuous-reward RL: selecting prompts by within-group reward variance rather than binary pass rate could benefit other dense-reward structured-generation tasks.
A controlled ablation separating DAPO's token-level importance ratios from its lack of KL anchoring would clarify whether the GSPO versus DAPO gap comes from update granularity or from the strength of regularization toward the SFT checkpoint.
The reported improvement in reading order is correlational; a direct ablation of bbox-based element matching against index-wise matching within the reward would test whether the matching mechanism drives that gain.

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

Read this as an honest system report rather than a claim of general Japanese document parsing capability. What is actually useful: a two-stream synthetic data engine with programmatic ground truth, a clear demonstration that mixed SFT recovers most VQA behavior lost under parsing-only SFT, and a thoughtful comparison of GSPO vs DAPO for this task. The thinking-mode vs instruct-mode SFT result (instruct mode hits 87.58 on parsing but crashes JA-Business-Doc-RQ-Bench to 0.384) is a genuine caution for anyone fine-tuning reasoning VLMs. The variance-based prompt filtering for continuous rewards also makes sense.

The soft spot is concentrated in the evaluation. OmniDocBench-JASyn is built by the same organization, uses synthetic images from a similar rendering pipeline, and the DocParse-Overall metric is a weighted average of text edit similarity, TEDS, and CDM—the same scoring functions used to build the RL reward in Table 3. The paper says the reward is "matched" to prevent proxy divergence, but it never addresses the fact that the benchmark is effectively a cousin of the training distribution. There is no holdout split, no real-document evaluation, no corruption or distribution-shift test. And the results are single runs: the 1.07-point DAPO improvement over mixed SFT could easily be noise, and the comparison in Table 8 has no error bars. The VQA benchmarks (JGraphQA-Refined, JDocQA-Refined, JA-Business-Doc-RQ-Bench) include some real-ish images, so the forgetting story is more convincing than the parsing capability story, but the central claim about parsing quality remains unverified outside the synthetic pipe.

The authors are careful about some things—they explicitly say the GSPO/DAPO comparison does not isolate a single design choice, and the reading-order result is labeled correlational. That honesty is real, and it makes the report more credible.

For a practitioner who wants a Japanese document parser and is willing to accept in-distribution performance, the released model is probably useful. For a researcher, the recipe is worth studying but the evaluation must be strengthened before the general-capability claim is taken seriously. I would send it to peer review, but with a request for external or at least independently held-out real-document evaluation, error bars or multiple seeds, and a sharper discussion of why reward/eval metric similarity is not itself harmless.

Referee Report

4 major / 5 minor

Summary. The paper post-trains Nemotron-3-Nano-Omni-30B-A3B-Reasoning for Japanese structured document parsing using a synthetic data engine with two streams: a Japanese Document VQA stream and a programmatic structured document parsing stream. It compares parsing-centric SFT, mixed SFT, and parsing-centric RL with GSPO and DAPO, and reports that mixed SFT preserves most VQA ability while matching parsing gains, and that DAPO-based parsing-centric RL further improves DocParse-Overall to 87.67 on OmniDocBench-JASyn. The released model is claimed to outperform several open-weight models on this benchmark. The paper also discusses reward design and variance-based prompt filtering for continuous structured parsing rewards.

Significance. If the central claim survives independent evaluation, the paper would offer a practical and transferable recipe: mixed SFT followed by DAPO-based parsing-centric RL can inject structured document parsing into a reasoning-oriented VLM while limiting VQA forgetting. The manuscript is clearly written, reports training hyperparameters, releases the final model, and gives a detailed description of a synthetic data engine with ground truth by construction. However, the main evidence is currently confined to a synthetic benchmark created by the same organization, and the evaluation metric overlaps substantially with the RL reward. The generality of the claimed parsing improvement is therefore unverified; the reported 87.67 may reflect in-distribution fit to the authors' own data generation and reward rather than a general document parsing capability.

major comments (4)

[§4.1.1, Table 3] The evaluation metric DocParse-Overall is the average of (1 − Text Edit)×100, Table TEDS, and Formula CDM, while the RL reward in Table 3 uses S_text = 1−NED, S_table = TEDS, and S_formula = CDM with positive weights. The headline metric is therefore directly optimized by the training reward. Because both the training data (§2.2) and the evaluation benchmark (§4.1.1, reference [6]) are synthetic and produced by the same organization, the reported improvement from 70.43 to 87.67 can be explained as in-distribution overfitting to the reward and the generator. The central claim of generalizable parsing improvement requires evaluation on a benchmark whose metric is not a reward component and whose pages are not produced by the same data engine, for example an independently constructed real-document corpus or a clearly disjoint distribution-shift set.
[§4.1.1, §2.2] No holdout split, contamination check, or provenance statement is reported. OmniDocBench-JASyn is released by the authors' organization and follows the same synthetic rendering approach as the training pipeline, but the paper never states that the 520 benchmark pages are disjoint from the training data generated by the engine in §2.2. Without explicit decontamination or an external evaluation, the comparison in Table 8 cannot rule out memorization of generator-specific layouts, rendering artifacts, or content distributions. The authors should provide a clear data-provenance table and evaluate on an independently created Japanese document benchmark.
[§4.4, Tables 6–7] All experiments are reported as single runs with no standard errors, no multiple seeds, and no significance testing. The key DAPO-over-Mixed-SFT gain is 1.07 DocParse-Overall points (86.60 to 87.67), and the Formula CDM gain is 3.39 points (89.48 to 92.87), while Text Edit is essentially unchanged. Given the variance of long-generation structured parsing outputs, these differences may be within noise. The authors should report multiple training runs or at least evaluation-level error bars and a paired significance test before claiming that DAPO improves beyond the SFT ceiling.
[§3.2.1, §3.2.2, §4.1.1] The reward includes components S_schema, S_bbox, and S_picture that do not appear in DocParse-Overall, and the weights in Table 3 are hand-chosen with no sensitivity analysis. The paper calls the reward 'matched' to the evaluation, but the overlap is only partial: text, table, and formula similarities are shared, while bounding-box and picture-description quality are optimized but not measured in the headline metric. This weakens the claim that the reward is application-aware and task-aligned. A weight-sensitivity study or an ablation that removes the unmatched components would clarify whether the reported gains come from the matched content terms or from the additional structure terms.

minor comments (5)

[Title and header] The title and running header contain spacing typos such as 'F orgetting' and 'Oﬀicial'; these should be corrected.
[§2.1, §2.2, §3.2.3, §4.2] The word 'diﬀicult' appears repeatedly instead of 'difficult'; a global typo pass is needed.
[Appendix A] The Japanese prompt in Appendix A appears as mojibake in the manuscript text; the appendix should be typeset with a correct Unicode font so the prompt is readable and reproducible.
[§3.2.3] The variance-filter thresholds σ_floor, μ_min, and μ_max are not reported, nor is the percentile-based calibration procedure described concretely; the authors should state the chosen values or the exact calibration rule.
[§4.4] The GSPO-versus-DAPO comparison intentionally varies multiple factors, and the paper acknowledges this; a single-factor ablation would strengthen the causal reading that DAPO's token-level updates or lack of KL anchoring drive the improvement.

Circularity Check

2 steps flagged · score 6.0 of 10

The claimed parsing improvement beyond the SFT ceiling is not an independent prediction: the RL reward is built from the same text/table/formula similarity functions as DocParse-Overall, and the only benchmark supporting the headline comparison is a same-organization synthetic dataset.

self definitional [Section 3.2.1 (Table 3) and Section 3.2.2 vs Section 4.1.1]
""DocParse-Overall = ((1 − Text Edit) × 100 + Table TEDS + Formula CDM)/3"; reward components: "S_text 0.45 Mean text similarity over all ground-truth text-class elements", "S_table 0.20 Mean TEDS over all ground-truth table elements", "S_formula 0.05 Mean formula similarity ... using CDM when available"; "This is designed to remove artificial order noise and to prevent proxy divergence between the training reward and the evaluation protocol.""

The reward's text, table, and formula sub-scores use the same normalized-edit-distance, TEDS, and CDM functions that define DocParse-Overall. The paper explicitly states that element matching is designed to prevent proxy divergence between the training reward and the evaluation protocol, confirming the reward is defined in terms of the evaluation metric. Therefore the RL stage directly optimizes the evaluation metric (with different weights and auxiliary bbox/schema/picture terms). The headline improvement from 86.60 to 87.67 is then a report of the training objective on in-distribution synthetic pages, not an independent prediction of general parsing quality. This makes the central 'beyond the SFT ceiling' claim self-definitional by construction.
self citation load bearing [Section 4.1.1 (reference [6]) and Section 4.5 (Table 8)]
""We evaluate structured document parsing performance on OmniDocBench-JASyn [6], which follows the OmniDocBench [7] format, a Japanese document parsing benchmark containing 520 synthetic Japanese document images covering 14 document types." "[6] Stockmark Inc. OmniDocBench-JASyn. https://huggingface.co/datasets/stockmark/OmniDocBench-JASyn, 2026." "Stockmark-Nemotron-3-Nano-Omni-JapanDocReader achieves a DocParse-Overall score of 87.67, outperforming the other open-weight models in this comparison.""

The central claim of superiority over open-weight models is supported exclusively by OmniDocBench-JASyn, a benchmark created by the same organization (Stockmark Inc.) that authored the paper and built the training-data engine. No holdout split, distribution-shift test, or real-document evaluation is reported. The benchmark pages and the training pages come from the same synthetic ecosystem, so the comparison in Table 8 measures fit to a self-created, in-distribution benchmark rather than independently verified document parsing ability. This is load-bearing self-citation: the headline result rests on a citation to the authors' own unverified dataset, and the reward-metric identity from Step 1 turns the evaluation into a self-referential loop.

full rationale

The SFT and forgetting-control comparisons are not inherently circular: the parsing-centric versus mixed SFT trade-off is an empirical observation measured on the same benchmark, and the VQA drift numbers on JA-Business-Doc-RQ-Bench, JGraphQA-Refined, and JDocQA-Refined have independent content. The circularity is concentrated in the final RL claim and its evaluation. The paper openly designs the parsing reward to match the evaluation protocol, using the same normalized edit distance, TEDS, and CDM components that constitute DocParse-Overall. Consequently, the DAPO-based improvement 'beyond the SFT ceiling' is an optimization of the test metric itself, and the reported 87.67 is a fitted quantity rather than a prediction. The external comparison is further weakened by the fact that OmniDocBench-JASyn is a same-organization synthetic benchmark with no reported disjointness from the training data engine and no real-document validation. These factors make the central result partially circular: the model is selected and optimized for a metric that is also the reward, and evaluated on a benchmark produced by the same pipeline. However, the paper does not hide the reward-metric alignment, and parts of the study (notably the forgetting analysis and the comparison between thinking and instruct modes) retain independent empirical content, so a score of 6 rather than 8 or 10 is appropriate.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central result rests on unshown assumptions about synthetic data realism, benchmark independence, and the hand-chosen RL/filter parameters. Data, code, thresholds, and real-document validation are not provided, so these assumptions cannot be checked from the paper alone.

free parameters (4)

Reward component weights = schema 0.15, text 0.45, table 0.20, formula 0.05, bbox 0.10, picture 0.05
Chosen by hand in Table 3, no sensitivity analysis; they determine what RL optimizes and hence the reported gain.
IoU matching threshold tau_IoU = unspecified
Used in greedy element matching for all reward sub-scores (Section 3.2.2); the paper says it is intentionally lenient but gives no numerical value.
Variance filter thresholds sigma_floor, mu_min, mu_max = unspecified
Section 3.2.3 uses these bounds to retain about 7k RL prompts; thresholds are described by type but not reported, and they shape which prompts are trained on.
Mixed SFT data mixing ratio = unspecified
The VQA pool is 94k and parsing pool is 25k, but the ratio used in mixed SFT is not stated; this ratio directly controls the forgetting/parsing trade-off reported in Table 6.

assumptions (4)

domain assumption Synthetic pages generated by the data engine and by OmniDocBench-JASyn are representative of real Japanese documents.
All parsing evaluation is on a Stockmark benchmark of 520 synthetic images; no real-document or distribution-shift test is reported (Section 4.1.1).
domain assumption The benchmark pages are disjoint from the training data and RL prompt pool.
The paper never states that OmniDocBench-JASyn was held out; both benchmark and training data come from the same group and use similar synthetic generation (Sections 3.1 and 4.1.1).
ad hoc to paper The hand-chosen reward weights and filtering thresholds define a valid task-level objective.
No sensitivity analysis or reported threshold values (Table 3, Section 3.2.3); the RL outcome depends on these choices.
domain assumption DocParse-Overall and VQA-Overall are faithful proxies for the claimed capabilities.
DocParse-Overall is an equally weighted mean of text edit, TEDS, and CDM; VQA-Overall averages three benchmarks. No human or downstream validation is reported (Section 4.1).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Stockmark-Nemotron-3-Nano-Omni-JapanDocReader: Structured Document Parsing via Capability Injection and Forgetting Control." pith.science (2026). https://pith.science/paper/TL7CGPZQ

@misc{pith2026260806758,
  author       = {Pith},
  title        = {Pith review of: Stockmark-Nemotron-3-Nano-Omni-JapanDocReader: Structured Document Parsing via Capability Injection and Forgetting Control},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TL7CGPZQ}},
  note         = {Machine review of arXiv:2608.06758}
}

read the original abstract

We present Stockmark-Nemotron-3-Nano-Omni-JapanDocReader, a Japanese document understanding model built from Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16. The central goal of this work is structured document parsing via capability injection and forgetting control: we inject Japanese structured document parsing capability into a reasoning-oriented multimodal model while preserving its document VQA capability as much as possible. We study parsing-centric SFT, which uses only structured document parsing data; mixed SFT, which combines structured document parsing and VQA data; and parsing-centric RL, which optimizes structured parsing with a task-level reward. Our experiments show that parsing-centric SFT substantially improves structured document parsing performance but causes measurable VQA forgetting. Mixed SFT mitigates this forgetting while preserving nearly the same structured parsing performance. Applying DAPO-based parsing-centric RL on top of the mixed SFT checkpoint further improves structured document parsing beyond the SFT ceiling, producing the final released model. The training data is constructed with a data engine consisting of two complementary synthetic streams: a Japanese Document VQA Stream and a programmatic structured document parsing stream. We also discuss reward design and variance-based prompt filtering for continuous structured document parsing rewards, highlighting their importance for making RL effective in long-reasoning structured document parsing tasks.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 11 canonical work pages

[6]

OmniDocBench-JASyn

Stockmark Inc. OmniDocBench-JASyn. https://huggingface.co/datasets/stockmar k/OmniDocBench-JASyn, 2026

work page 2026
[1]

Nemotron 3 Nano Omni: Eﬀicient and open multimodal intelligence

Amala Sanjay Deshmukh, Kateryna Chumachenko, Tuomas Rintamaki, Matthieu Le, Tyler Poon, Danial Mohseni Taheri, Ilia Karmanov, Guilin Liu, Jarno Seppanen, Arushi Goel, et al. Nemotron 3 Nano Omni: Eﬀicient and open multimodal intelligence. arXiv preprint arXiv:2604.24954, 2026

arXiv 2026
[2]

Nemotron-Personas-Japan: Synthetic personas aligned to real- world distributions

Atsunori Fujita, Vincent Gong, Masaya Ogushi, Kotaro Yamamoto, Yoshi Suhara, Dane Corneil, and Yev Meyer. Nemotron-Personas-Japan: Synthetic personas aligned to real- world distributions. https://huggingface.co/datasets/nvidia/Nemotron-Personas-J apan, September 2025

work page 2025
[3]

Qwen3.6-27B: Flagship-level coding in a 27B dense model

Qwen Team. Qwen3.6-27B: Flagship-level coding in a 27B dense model. https://qwen.a i/blog?id=qwen3.6-27b, April 2026

work page 2026
[4]

Group sequence policy optimization

Chujie Zheng, Shixuan Liu, Mingze Li, Xiong-Hui Chen, Bowen Yu, Chang Gao, Kai Dang, Yuqiong Liu, Rui Men, An Yang, et al. Group sequence policy optimization. arXiv preprint arXiv:2507.18071, 2025

arXiv 2025
[5]

DAPO: An open-source LLM reinforcement learning system at scale

Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Weinan Dai, Tiantian Fan, Gaohong Liu, Lingjun Liu, et al. DAPO: An open-source LLM reinforcement learning system at scale. In Advances in Neural Information Processing Systems , volume 38, pages 113222–113244, 2025

work page 2025
[7]

OmniDocBench: Benchmarking diverse PDF document parsing with comprehensive annotations

Linke Ouyang, Yuan Qu, Hongbin Zhou, Jiawei Zhu, Rui Zhang, Qunshu Lin, Bin Wang, Zhiyuan Zhao, Man Jiang, Xiaomeng Zhao, et al. OmniDocBench: Benchmarking diverse PDF document parsing with comprehensive annotations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 24838–24848, 2025

work page 2025
[8]

Binary codes capable of correcting deletions, insertions, and reversals

Vladimir I Levenshtein et al. Binary codes capable of correcting deletions, insertions, and reversals. Soviet Physics Doklady , 10(8):707–710, 1966

work page 1966

Show all 16 references

[9]

Image-based table recognition: data, model, and evaluation

Xu Zhong, Elaheh ShafieiBavani, and Antonio Jimeno Yepes. Image-based table recognition: data, model, and evaluation. In European Conference on Computer Vision , pages 564–580. Springer, 2020

2020
[10]

CDM: A reliable metric for fair and accurate formula recognition evalua- tion

Bin Wang, Fan Wu, Linke Ouyang, Zhuangcheng Gu, Rui Zhang, Renqiu Xia, Bo Zhang, and Conghui He. CDM: A reliable metric for fair and accurate formula recognition evalua- tion. arXiv preprint arXiv:2409.03643 , 2024

2024 arXiv
[11]

JA-Business-Doc-RQ-Bench

Stockmark Inc. JA-Business-Doc-RQ-Bench. https://huggingface.co/datasets/stoc kmark/JA-Business-Doc-RQ-Bench , 2026

2026
[12]

JGraphQA

Akira Kinoshita. JGraphQA. https://huggingface.co/datasets/r-g2-2024/JGraphQA, 2025. 18

2024
[13]

JAMMEval: A refined collection of Japanese benchmarks for reliable VLM eval- uation

Issa Sugiura, Koki Maeda, Shuhei Kurita, Yusuke Oda, Daisuke Kawahara, and Naoaki Okazaki. JAMMEval: A refined collection of Japanese benchmarks for reliable VLM eval- uation. arXiv preprint arXiv:2604.00909 , 2026

2026 arXiv
[14]

JDocQA: Japanese doc- ument question answering dataset for generative language models

Eri Onami, Shuhei Kurita, Taiki Miyanishi, and Taro Watanabe. JDocQA: Japanese doc- ument question answering dataset for generative language models. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-...

2024
[15]

Qwen3.5: Towards native multimodal agents

Qwen Team. Qwen3.5: Towards native multimodal agents. https://qwen.ai/blog?id= qwen3.5, February 2026

2026
[16]

document_structure

Gemma Team. Gemma 4 technical report, 2026. URL https://arxiv.org/abs/2607.0 2770. 19 Appendix A. Structured Document Parsing Prompt For structured document parsing, we use a fixed Japanese prompt across evaluation and model comparison. The prompt asks the model to extract the...

2026

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [6]

OmniDocBench-JASyn

Stockmark Inc. OmniDocBench-JASyn. https://huggingface.co/datasets/stockmar k/OmniDocBench-JASyn, 2026

work page 2026

[2] [1]

Nemotron 3 Nano Omni: Eﬀicient and open multimodal intelligence

Amala Sanjay Deshmukh, Kateryna Chumachenko, Tuomas Rintamaki, Matthieu Le, Tyler Poon, Danial Mohseni Taheri, Ilia Karmanov, Guilin Liu, Jarno Seppanen, Arushi Goel, et al. Nemotron 3 Nano Omni: Eﬀicient and open multimodal intelligence. arXiv preprint arXiv:2604.24954, 2026

arXiv 2026

[3] [2]

Nemotron-Personas-Japan: Synthetic personas aligned to real- world distributions

Atsunori Fujita, Vincent Gong, Masaya Ogushi, Kotaro Yamamoto, Yoshi Suhara, Dane Corneil, and Yev Meyer. Nemotron-Personas-Japan: Synthetic personas aligned to real- world distributions. https://huggingface.co/datasets/nvidia/Nemotron-Personas-J apan, September 2025

work page 2025

[4] [3]

Qwen3.6-27B: Flagship-level coding in a 27B dense model

Qwen Team. Qwen3.6-27B: Flagship-level coding in a 27B dense model. https://qwen.a i/blog?id=qwen3.6-27b, April 2026

work page 2026

[5] [4]

Group sequence policy optimization

Chujie Zheng, Shixuan Liu, Mingze Li, Xiong-Hui Chen, Bowen Yu, Chang Gao, Kai Dang, Yuqiong Liu, Rui Men, An Yang, et al. Group sequence policy optimization. arXiv preprint arXiv:2507.18071, 2025

arXiv 2025

[6] [5]

DAPO: An open-source LLM reinforcement learning system at scale

Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Weinan Dai, Tiantian Fan, Gaohong Liu, Lingjun Liu, et al. DAPO: An open-source LLM reinforcement learning system at scale. In Advances in Neural Information Processing Systems , volume 38, pages 113222–113244, 2025

work page 2025

[7] [7]

OmniDocBench: Benchmarking diverse PDF document parsing with comprehensive annotations

Linke Ouyang, Yuan Qu, Hongbin Zhou, Jiawei Zhu, Rui Zhang, Qunshu Lin, Bin Wang, Zhiyuan Zhao, Man Jiang, Xiaomeng Zhao, et al. OmniDocBench: Benchmarking diverse PDF document parsing with comprehensive annotations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 24838–24848, 2025

work page 2025

[8] [8]

Binary codes capable of correcting deletions, insertions, and reversals

Vladimir I Levenshtein et al. Binary codes capable of correcting deletions, insertions, and reversals. Soviet Physics Doklady , 10(8):707–710, 1966

work page 1966