AI Open Innovation Day 2024
2024-05-15
Tokyo, Japan
who it's for
LLM engineers, research scientists, open source maintainers, MLOps engineers, data engineers, AI product managers
topics
open source LLMs and checkpoints; instruction tuning pipelines; synthetic training corpora generation; dataset curation and filtering; Japanese corpus preprocessing; vocabulary expansion for context; model merge techniques; experiment tracking and MLOps tooling; model leaderboard evaluation; large-scale training infrastructure (ABCI, GPU support); annotation workforce and labeling pipelines; safety and governance for open AI
| title | author | length |
|---|---|---|
| Keynote Panel: Open Innovation in AI - Present an... | Christopher Dewan, Mariko Isogawa, Miori Sagara | 51m |
Research groups demonstrate estimating full human pose from sparse sensors such as laser reflections, event cameras, and even acoustic-only inputs, and use… | ||
| Welcoming Remarks | Noly Morimoto | 04m |
Morning welcome summarizes AI progress with concrete milestones: Deep Blue vs Kasparov in the 1990s and Watson on Jeopardy 15 years later, claiming the system… | ||
| AI Open Innovation and AI Alliance | Anthony J. Annunziata | 32m |
Announces an alliance now above 100 members co-launched with Meta to build open AI infrastructure and governance, organized into six focus areas including… | ||
| Keynote: Foundation to Build LLMs | Naoaki Okazaki | 42m |
Swallow is a family of LLMs created by continuing pretraining on Llama 2 and Mistral bases using compute on ABCI. The training used a cleaned Japanese corpus… | ||
| Keynote: LLM Dev Status in Japan | Akira Shibata | 42m |
Weights & Biases presents concrete tooling for experiment tracking and MLOps with weights and biases used by tens of thousands in Japan (20–25k) and Korea… | ||
| Keynote: LLM Research and Development at National Institute of Informatics | Sadao Kurohashi | 30m |
LM-JP builds fully open models with training code, checkpoints and a searchable training corpus; the first release is LJP-13B (GPT-2 style, 13B params) trained… | ||
| nothing matches | ||