speed1313
RSS FeedI am a first year PhD student at Institute of Science Tokyo, majoring in Computer Science. I like to learn by building things.
Posts
-
Pitfalls in Model Evaluation
Model evaluation is full of pitfalls. This post walks through the many factors that affect evaluation scores.
- Benchmarks
-
My Trip to IHPCSS2026 in Perth
A report on attending IHPCSS2026, a summer school on HPC held in Perth, Australia.
- Events
-
HakushoBench: Building a Japanese Chart and Table VQA Benchmark from White Papers
This post introduces HakushoBench, a new Japanese chart and table VQA benchmark that we built.
- Benchmarks
-
Jagle: Building a Large-Scale Japanese Multimodal Post-Training Dataset for Vision-Language Models
This post introduces Jagle, a new Japanese multimodal post-training dataset of about 9.2M examples that we built.
- Resources
-
My Trip to ICLR 2026 in Rio
A record of my trip to ICLR 2026, an international conference held in Rio de Janeiro.
- Events
-
Introducing LLM-jp-4-VL-9B-beta, a Vision-Language Model
This post introduces LLM-jp-4-VL-9B-beta (beta release), a vision-language model developed by LLM-jp.
- Models
-
Introducing simple-evals-mm
This post introduces simple-evals-mm, a new VLM evaluation framework we have developed.
- Benchmarks
-
JAMMEval: Refining Japanese Benchmarks for Reliable VLM Evaluation
This post introduces JAMMEval, a benchmark collection built by manually refining seven existing Japanese VQA benchmarks.
- Benchmarks
-
My Favorite Spots and Food in Kyoto
Recommended spots and food I discovered in Kyoto, where I spent two years as a master's student.
- Life
-
Recommended Resources for Research
This post introduces recommended resources that are useful for research.
- Tips
-
An Introduction to Terminal-Bench & Harbor
This post explains Terminal-Bench, one of the most important benchmarks today, and Harbor, the evaluation framework behind it.
- Benchmarks
-
WAON: A Large-Scale, High-Quality Japanese Image-Text Pair Dataset
This post introduces WAON, a large-scale, high-quality Japanese image-text pair dataset that we built at LLM-jp.
- Resources
-
Introducing Llama-Mimi, a Speech Language Model
This post introduces Llama-Mimi, a speech language model developed at LLM-jp.
- Models
-
My Trip to NAACL 2025 in Albuquerque
A record of my attendance at NAACL 2025, an international NLP conference held in Albuquerque.
- Events
-
Introducing llm-jp-modernbert
An introduction to llm-jp-modernbert
- Models
-
Recommended Papers for Foundation Model Research
This post lists papers that are good to read first when starting foundation model research.
- Tips
-
Developing Japanese CLIP Models Leveraging an Open-weight LLM for Large-scale Dataset Translation
This post introduces our work on building a large-scale Japanese image-text pair dataset with an open LLM and using it to develop a Japanese CLIP model.
- Models
-
Introducing llm-jp-eval-mm
This post introduces llm-jp-eval-mm, a tool for evaluating the Japanese-language performance of vision-language models, which I am developing together with Silviase.
- Benchmarks
-
text2dataset: A Tool for Translating Datasets with Open LLMs
This post introduces text2dataset, a tool that uses open LLMs to build Japanese translated datasets from English datasets.
- Resources
-
Let's Apply The AI Scientist to the SEIR Model
Let's apply The AI Scientist to the SEIR model, a differential equation-based model used to predict the spread of infectious diseases.
- Agents
-
Introduction to MinHash
This post explains the basic idea and algorithm of MinHash, which is widely used for deduplication when building corpora for large language models, and how to reduce the number of comparisons.
- Notes
-
Pretraining an LLM from Scratch
To fully understand LLM pretraining, I pretrained an LLM entirely from scratch in JAX. This post describes how.
- Models
-
Python's Garbage Collection
In this post, we will examine the garbage collection (GC) in CPython. Furthermore, we will also look into the relationship between GC and Global Interpreter Lock (GIL).
- Notes
-
The Appeal of Osaka University's Main Library
In this post, I introduce the appeal of the Osaka University Main Library that I discovered during my undergraduate years.
- Life
-
An Introduction to the Neural Tangent Kernel, with Experiments
This post introduces Neural Tangent Kernel (NTK) theory, which underpins research focused on the dynamics of training.
- Notes
-
My Graduate School Entrance Exams in 2023
Hi, I'm speed. In the summer of 2023, called the hottest summer on record amid relentless heat, I took entrance exams for master's programs. With computer science so popular, I expected getting in to be hard, so I applied to several graduate schools.
- Life
-
Optimal Transport
This post explains the basic ideas of optimal transport and some of its applications.
- Notes
-
How Building My Own OS Changed the Way I See Operating Systems
In this post, I describe how building my own OS changed the way I see operating systems.
- Notes
-
Garbage Collection: The Mark & Sweep Method
This post explains the mark & sweep method, one of the algorithms used for garbage collection.
- Notes
-
The World of Computer Science: A Letter to My First-Year Self
In this post, written for my first-year university self who was a stray sheep, I lay down a “highway of knowledge”: I explain how a browser works from the bottom up and, along the way, list good books on each topic to show how to approach learning computer science.
- Tips
-
My Experience at Security Camp 2022, Track B (Web Security)
I recently took part in Security Camp 2022, Track B (Web Security).
- Events
No posts with these tags.