Presentation materials and surveys used during the study sessions are available to the public. Note that the contents are mainly written in Japanese.
Additionally, we are updating the Overview of Japanese LLMs.
2025-4-22
- LLM-jp Status Report (Kurohashi) [PDF]
- Harnessing AI Agents for Real-World Applications ( Kwasi Ankomah / SambaNova, Principle AI Engineer）[PDF]
- The role of LLM in juris-informatics and a survey of legal control over LLM (Sato) [PDF]
- mid-training for LLM [PDF]
- Dialogue WG (Higashinaka) [PDF]
- Corpus Construction Working Group (Kawahara) [PDF]
- Real-world Interaction Working Group (Ogata) [PDF]
- Evaluation and Tuning Working Group (Miyao) [PDF]
- Safety Working Group (Sekine) [PDF]
- Multi-modal Working Group (Okazaki) [PDF]
- Model Building Working Group (Suzuki) [PDF]
2025-3-25
- LLM-jp Status Report (Kurohashi) Oral
＜Evaluation and Turning/Principal Elucidation WG＞
- Are Checklists Really Useful for Automatic Evaluation of Generative Tasks? (Furuhashi) [PDF]
- Introduction of Open Japanese LLM leaderboard and statistical analysis on evaluation results. (Namgi Han)[PDF]
- Analyzing the Pretraining of Japanese Large Language Models. (Nishida) [PDF]
- llm-jp-judge: Japanese LLM-as-a-Judge Evaluation Tool. (Kodama) [PDF]
- Understanding the Role of Persona and Internal Mechanisms in Large Language Models. (Ozaki)[PDF]
- How LLMs Learn: Tracing Internal Representations with Sparse Autoencoders. (Inaba)[PDF]
- A Massive Fine-tuned LLMs from Diverse Models, Tasks, Methods (Harada) [PDF]
- Comparative analysis of the Geospatial Representations in Large Language Models across Models and Languages (Otake) [PDF]
- Large-Scale Human Evaluation of LLMs for Japanese（Inoue） [PDF]
- A Study on Fine-tuning Methods for Balancing Usefulness and Safety in Japanese Large Language Models. (Katsumata)[PDF]
＜Multi-modal WG＞
- Developing Japanese CLIP Models Leveraging an Open-weight LLM for Large-scale Dataset Translation. (Sugiura) [PDF]
- lm-jp-eval-mm: An Evaluation Framework for Evaluating Japanese-centric Vision and Language Model. (Sugiura) [PDF]
- LLM-jp-3 VILA: Construction of Japanese Multimodal Data and Powerful Japanese Multimodal Model (Sasagawa) [PDF]
＜Model Building WG＞
- Drop-Upcycling: Training Sparse Mixture of Experts with Partial Re-initialization (Nakamura) [PDF]
＜Safety WG＞
- Large-Scale Human Evaluation of LLM Safety (Takahashi) [PDF]
- AnswerCarefully: AnswerCarefully: A Dataset for Promoting Safety of Japanese LLMs (Suzuki)[PDF]
- Developing a Dataset of Misinformation from Social Media and an Accuracy Benchmark for Large Language Models (Nakazato)[PDF]
- Development of Prompt Attack Data Collection Application for LLMs and Analysis of Collected Data Characteristics (Hayashi)[PDF]
＜Corpus Construction WG＞
- A Comprehensive Analysis of Memorization in Large Language Models (Kiyomaru) [PDF]
- Detection of Sensitive Personal Information in the Pre-training Corpus for Large Language Models (Minamoto) [PDF]
- Integrated Framework for LLM Domain Adaptation Based on Synthetic Data (Ogawa) [PDF]
2025-2-25
- LLM-jp Status Report (Kurohashi) Oral
- Real-world Interaction Working Group (Ogata) [PDF]
- Safety Working Group (Sekine) [PDF]
- Japan AI Safety Institute (Semitsu) [PDF]
- Model Building Working Group (Suzuki) [PDF]
- Multi-modal Working Group (Okazaki) [PDF]
- Corpus Construction Working Group (Kawahara) [PDF]
- Evaluation and Tuning Working Group (Miyao) [PDF]
- Dataflow Architecture Achieving 198 Tokens per Second with DeepSeek R1 671B (Kenichi Hayashi/SambaNova System) [PDF]
- PLaMo 2 Tokenizer: Keys to Token Efficiency (Kentaro Imajo/Preferred Networks) [PDF]
- Training progress of LLM-jp-3 models: analysis on downstream performance (Nishida/Oda) [PDF]
2025-1-16
- LLM-jp Status Report (Kurohashi) [PDF]
- Model Building Working Group (Suzuki) [PDF]
- Multi-modal Working Group (Okazaki) [PDF]
- Evaluation and Tuning Working Group (Miyao) [PDF]
- Real-world Interaction Working Group (Ogata) [PDF]
- Corpus Construction Working Group (Kawahara) [PDF]
- Junichiro Takahashi/The University of Tokyo [PDF]
- Kaito Baba/The University of Tokyo [PDF]
- Efforts to Efficiently Create High-Quality LLM Datasets (Yujiro Terazawa/APTO,Inc) [PDF]
- LLM Training Using Synthetic Data (Kiyomaru) [PDF]
2024-11-26
- LLM-jp Status Report (Kurohashi) Oral Presenation
- Development and Evaluation of Tanuki (Kan Hatakeyama/Institute of Science Tokyo) [PDF]
- EMNLP2024 Report (Takagi) [PDF] (Kodama) [PDF] (Liu) [PDF]
- Safety Working Group (Sekine) [PDF]
- Corpus Construction Working Group (Kawahara) [PDF]
- Evaluation and Tuning Working Group (Miyao) [PDF]
- Real-world Interaction Working Group (Ogata) [PDF]
- Model Building Working Group (Suzuki) [PDF]
- Multi-modal Working Group (Okazaki) [PDF]
- Discussion on the Training Progress of LLM-jp-3 172B (Oda) [PDF]
2024-10-29
- LLM-jp Status Report (Kurohashi)
- BritLLM:Organising, producing, and publishing the first British Large Language Model (Pontus Stenetorp/NII)
- Pre-training and Post-training of PLaMo100B (Hiroaki Mikami,Kosuke Nakago/Preferred Elements)
- Multi-modal Working Group (Okazaki,Sasagawa,Maeda,Sugiura)
- Model Building Working Group (Suzuki)
- Corpus Construction Working Group (Kawahara)
- Evaluation and Tuning Working Group (Miyao)
- Safety Working Group (Sekine)
- Real-world Interaction Working Group (Ogata)
2024-08-27
- On Web Article Crawling and Copyright Infringement (Kakinuma)
- Beyond English-Centric LLMs: What Language Do Multilingual Language Models Think in? (Qianying Liu/NII)
- Real-world Interaction Working Group (Ogata)
- Corpus Construction Working Group (Kawahara)
- Evaluation and Tuning Working Group (Miyao)
- Safety Working Group (Sekine)
- Multi-modal Working Group (Okazaki)
- Model Building Working Group (Suzuki)
2024-07-30
- LLM-jp Status Report (Kurohashi)
- What is Open Source AI? Explaining the Draft Version of “The Open Source AI Definition” (Sado)
- Recent Advances in Addressing Hallucinations (Tsuta)
- Corpus Construction Working Group (Kawahara)
- Safety Working Group (Sekine)
- Evaluation and Tuning Working Group (Miyao)
- Multi-modal Working Group (Okazaki)
- Report on the Setup of Sakura Cluster (Kuga)
- Model Building Working Group (Suzuki)
2024-06-25
- LLM-jp Status Report (Kurohashi)
- Latest Advancements in Document Image Understanding with Large Language Models (Tanaka)
- Development of Nejumi Leaderboard3 (Kamata)
- Sarashina: Introduction to Japanese LLMs developed by SB Intuitions (Takase)
- Mechanistic Interpretability: Introduction to Scaling Monosemanticity (Anthropic, 2024) (Takagi)
- Corpus Construction Working Group (Kawahara)
- Model Building Working Group (Suzuki)
- Evaluation and Tuning Working Group (Miyao)
- Safety Working Group (Sekine)
- Multi-modal Working Group (Okazaki)
2024-05-26
- LLM-jp Status Report (Kurohashi)
- Development of a Japanese LLM with 100B Parameters (Omi)
- Overview of R&D for Generative AI at Databricks (Yayoi)
- Multilingual LLM: Data Construction, Fine-tuning, and LLM-based Evaluation (Peter)
- Corpus Construction Working Group (Kawahara)
- Model Building Working Group (Suzuki)
- Evaluation and Tuning Working Group (Miyao)
- Safety Working Group (Sekine)
- Multi-modal Working Group (Okazaki)
2024-03-26
- LLM-jp Status Report (Kurohashi)
- Geniac Program Initiatives at ABEJA (Hattori)
- Estimated timing of transition from Prompt Tuning to Fine Tuning (Kubo)
- An experiment to reduce halucinations produced by ichikara-instruction, Less NE experiments (Sekine)
- Evaluation and Tuning Working Group (Miyao)
- Corpus Construction Working Group (Kawahara)
- Safety Working Group (Sekine)
- Model Building Working Group (Suzuki)
2024-01-22
- Self-improvement loops from observational data, Grounding LLMs with causal inference methods (Sannai)
- Comparative analysis of LLM human evaluation and GPT evaluation using ichikara-instruction (Sekine)
- Kotoba Technologies voice-based model and Mamba architecture (Kojima and Kasai)
- Large language model, Swallow (TokyoTech LLM members)
- Safety Working Group (Sekine)
- Corpus Construction Working Group (Kawahara)
- Evaluation and Tuning Working Group (Miyao)
- Model Building Working Group (Suzuki)
2023-11-29
- LLM-jp Status Report (Kurohashi)
- LLM Security Survey and Japanese Data (Suzuki)
- Development of 13 Billion Parameter Japanese Pre-Learning Model for Business Domains and Latest Information (Omi)
- Potential and Progress of Large-scale Language Models in Medicine (Kodera)
- Corpus Construction Working Group (Kawahara)
- Model Building Working Group (Suzuki)
- Evaluation and Tuning Working Group (Miyao)
- Safety Working Group (Sekine)
2023-10-18
- Training and applying Vision & Language model with Heron (Inoue)
- Technical overview of Japanese LLM “ELYZA-japanese-Llama-2-7b” based on “Llama 2” from Meta (Nakamura)
- Technical overview of pre-training for PLaMo-13B (Mikami)
- LLM models for robotics (Kawaharazuka)
2023-09-04
- A survey for instruction tuning for Model imitation explained (Mizuki)
- PEFT: LazyLoRA (Wu)
- Overview of R&D for LLM at Stability AI Japan (Lee)
- Generative AI, construction and copyright (Kakinuma)
- Construction of a large-scale bilingual English/Japanese language model “Weblab-10B” (Kojima)
2023-07-20
- ACL2023 participation announcement (Kodama) (Yamada) (Ueda) (Deguchi)
- LLM Peripheral technology (June/July 2023) (Tsuruoka)
2023-06-19
- Overview of the model project for CyberAgent (Ishigami)
- Tips for training of T5 in Japanese (Nishitoba)
- Participation report for the ABCI Grand challenge (Sakaguchi)
- Research and development of large-scale language models at NICT (Torizawa)
- Future Cooperation for NII：Domain adaptation for Biomedical (Aizawa)