Yuu Jinnai


Home
Google Scholar
GitHub

Current Research:
Language Model Finetuning
Minimum Bayes Risk Decoding
Language Model Alignment

Prior Projects:
Parallel Best-First Search
Automated Skill Discovery

日本語:
Japanese
Open Data Structures
ヒューリスティック探索入門


Hosted on GitHub Pages — Theme by orderedlist

Language Model Finetuning

Language model finetuning and post-training specialize pre-trained foundation models into efficient, robust, and capable models for target tasks. My work focuses on developing compact domain-specialized models, reasoning-oriented models for non-English languages, and effective alignment and preference optimization methods.

All open-source models and datasets are available on Hugging Face (@cyberagent).


Developed Models

Reasoning Models (RLMs)

cost-of-reasoning

Model Base Architecture Description Links
CAT-Thinking-8B Qwen3-8B Japanese reasoning model producing explicit Japanese thinking traces. Addresses the performance disparity and “reasoning tax” when solving non-English reasoning tasks. MODEL | PAPER | DATASET
CAT-Paws-8B Qwen3-8B Sibling reasoning model of CAT-Thinking-8B specially tuned for agentic, multi-turn, and tool-use environments. MODEL | PAPER | DATASET

Compact Machine Translation (CAT-Translate Series)

cat-translate

The CAT-Translate series demonstrates that compact, bilingual-specialized language models can match or outperform large multilingual models while providing significantly lower inference latency and memory requirements.

Model Parameters Base Model Links
CAT-Translate-7b 7B Mistral MODEL | PAPER | DATASET
CAT-Translate-3.3b 3.3B Llama-3.2 MODEL | PAPER | DATASET
CAT-Translate-1.4b 1.4B Qwen2.5 MODEL | PAPER | DATASET
CAT-Translate-0.8b 0.8B Qwen2.5 MODEL | PAPER | DATASET

Preference Optimization & Self-Improvement

Model Base Architecture Description Links
calm3-22b-chat-selfimprove-experimental calm3-22b-chat Self-improvement conversational model leveraging calm3-22b-chat for data augmentation and self-training. MODEL
calm2-7b-chat-dpo-experimental calm2-7b Conversational Japanese LLM aligned with Direct Preference Optimization (DPO). Investigates cross-cultural alignment and commonsense morality. MODEL | PAPER | DATASET

Datasets and Benchmarks

We have developed and released multiple public datasets and evaluation benchmarks on Hugging Face:


Associated Publications

  1. Jinnai Y. 2027. JOR-Bench: Japanese Operations Research Benchmarks for Large Language Models. In Findings of the Association for Computational Linguistics (EACL-27 Findings).
    PAPER | DATASET
  2. Jinnai Y. 2026. Cost of Reasoning in non-English Languages: A Case Study on Japanese. arXiv preprint arXiv:2607.10114.
    PAPER | MODEL (Thinking) | MODEL (Paws) | DATASET
  3. Jinnai Y. 2026. CAT-Translate: Building Compact Open-Source Models for Japanese-English Translation. arXiv preprint arXiv:2606.21413.
    PAPER | MODELS | DATASET
  4. Ichihara Y, Jinnai Y, Ariu K, Uchibe E. 2026. Consensus Group Relative Policy Optimization for Text Generation. Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP-26 Main).
    PAPER | CODE
  5. Ichihara Y, Jinnai Y, Morimura T, Sakamoto M, Mitsuhashi R, Uchibe E. 2026. MO-GRPO: Mitigating Reward Hacking of Group Relative Policy Optimization on Multi-Objective Problems. Transactions of the Association for Computational Linguistics (TACL 2026).
    PAPER | CODE
  6. Ichihara Y, Jinnai Y. 2025. Auto-Weighted Group Relative Preference Optimization for Multi-Objective Text Generation Tasks. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: Industry Track (EMNLP-25 Industry).
    PAPER | CODE
  7. Jinnai Y, Honda U. 2025. Annotation-Efficient Preference Optimization for Language Model Alignment. Findings of the Association for Computational Linguistics (EMNLP-25 Findings).
    PAPER | CODE
  8. Jinnai Y. 2024. Does Cross-Cultural Alignment Change the Commonsense Morality of Language Models? Proceedings of the 2nd Workshop on Cross-Cultural Considerations in NLP (C3NLP Workshop at ACL 2024). 🏆 Best Paper Award
    PAPER | MODEL | DATASET