speed1313 Blog.https://speed1313.github.io/https://speed1313.github.io/en/posts/pitfalls-in-evaluation/https://speed1313.github.io/en/posts/pitfalls-in-evaluation/Model evaluation is full of pitfalls. This post walks through the many factors that affect evaluation scores.Wed, 19 Aug 2026 00:00:00 GMThttps://speed1313.github.io/en/posts/ihpcss2026/https://speed1313.github.io/en/posts/ihpcss2026/A report on attending IHPCSS2026, a summer school on HPC held in Perth, Australia.Mon, 20 Jul 2026 00:00:00 GMThttps://speed1313.github.io/en/posts/HakushoBench/https://speed1313.github.io/en/posts/HakushoBench/This post introduces HakushoBench, a new Japanese chart and table VQA benchmark that we built.Tue, 02 Jun 2026 00:00:00 GMThttps://speed1313.github.io/en/posts/Jagle/https://speed1313.github.io/en/posts/Jagle/This post introduces Jagle, a new Japanese multimodal post-training dataset of about 9.2M examples that we built.Tue, 12 May 2026 00:00:00 GMThttps://speed1313.github.io/en/posts/iclr2026/https://speed1313.github.io/en/posts/iclr2026/A record of my trip to ICLR 2026, an international conference held in Rio de Janeiro.Fri, 01 May 2026 00:00:00 GMThttps://speed1313.github.io/en/posts/LLM-jp-4-VL-9B-beta/https://speed1313.github.io/en/posts/LLM-jp-4-VL-9B-beta/This post introduces LLM-jp-4-VL-9B-beta (beta release), a vision-language model developed by LLM-jp.Tue, 14 Apr 2026 00:00:00 GMThttps://speed1313.github.io/en/posts/simple-evals-mm/https://speed1313.github.io/en/posts/simple-evals-mm/This post introduces simple-evals-mm, a new VLM evaluation framework we have developed.Tue, 07 Apr 2026 00:00:00 GMThttps://speed1313.github.io/en/posts/JAMMEval/https://speed1313.github.io/en/posts/JAMMEval/This post introduces JAMMEval, a benchmark collection built by manually refining seven existing Japanese VQA benchmarks.Sun, 05 Apr 2026 00:00:00 GMThttps://speed1313.github.io/en/posts/kyoto-osusume/https://speed1313.github.io/en/posts/kyoto-osusume/Recommended spots and food I discovered in Kyoto, where I spent two years as a master's student.Fri, 27 Mar 2026 00:00:00 GMThttps://speed1313.github.io/en/posts/research-resources/https://speed1313.github.io/en/posts/research-resources/This post introduces recommended resources that are useful for research.Fri, 06 Mar 2026 00:00:00 GMThttps://speed1313.github.io/en/posts/terminal-bench/https://speed1313.github.io/en/posts/terminal-bench/This post explains Terminal-Bench, one of the most important benchmarks today, and Harbor, the evaluation framework behind it.Sun, 08 Feb 2026 00:00:00 GMThttps://speed1313.github.io/en/posts/WAON/https://speed1313.github.io/en/posts/WAON/This post introduces WAON, a large-scale, high-quality Japanese image-text pair dataset that we built at LLM-jp.Tue, 28 Oct 2025 00:00:00 GMThttps://speed1313.github.io/en/posts/llama-mimi/https://speed1313.github.io/en/posts/llama-mimi/This post introduces Llama-Mimi, a speech language model developed at LLM-jp.Sun, 21 Sep 2025 00:00:00 GMThttps://speed1313.github.io/en/posts/naacl2025/https://speed1313.github.io/en/posts/naacl2025/A record of my attendance at NAACL 2025, an international NLP conference held in Albuquerque.Mon, 05 May 2025 00:00:00 GMThttps://speed1313.github.io/en/posts/llm-jp-modernbert/https://speed1313.github.io/en/posts/llm-jp-modernbert/An introduction to llm-jp-modernbertWed, 23 Apr 2025 00:00:00 GMThttps://speed1313.github.io/en/posts/paper-list-foundation-model/https://speed1313.github.io/en/posts/paper-list-foundation-model/This post lists papers that are good to read first when starting foundation model research.Sun, 20 Apr 2025 00:00:00 GMThttps://speed1313.github.io/en/posts/llm-jp-clip/https://speed1313.github.io/en/posts/llm-jp-clip/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.Wed, 12 Feb 2025 00:00:00 GMThttps://speed1313.github.io/en/posts/llm-jp-eval-mm/https://speed1313.github.io/en/posts/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.Thu, 06 Feb 2025 00:00:00 GMThttps://speed1313.github.io/en/posts/text2dataset/https://speed1313.github.io/en/posts/text2dataset/This post introduces text2dataset, a tool that uses open LLMs to build Japanese translated datasets from English datasets.Tue, 29 Oct 2024 00:00:00 GMThttps://speed1313.github.io/en/posts/ai_scientist/https://speed1313.github.io/en/posts/ai_scientist/Let's apply The AI Scientist to the SEIR model, a differential equation-based model used to predict the spread of infectious diseases.Wed, 09 Oct 2024 00:00:00 GMThttps://speed1313.github.io/en/posts/minhash/https://speed1313.github.io/en/posts/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.Sun, 14 Jul 2024 00:00:00 GMThttps://speed1313.github.io/en/posts/llm-from-scratch/https://speed1313.github.io/en/posts/llm-from-scratch/To fully understand LLM pretraining, I pretrained an LLM entirely from scratch in JAX. This post describes how.Fri, 14 Jun 2024 00:00:00 GMThttps://speed1313.github.io/en/posts/python-gc/https://speed1313.github.io/en/posts/python-gc/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).Thu, 11 Apr 2024 00:00:00 GMThttps://speed1313.github.io/en/posts/handai_library/https://speed1313.github.io/en/posts/handai_library/In this post, I introduce the appeal of the Osaka University Main Library that I discovered during my undergraduate years.Fri, 23 Feb 2024 00:00:00 GMThttps://speed1313.github.io/en/posts/ntk/https://speed1313.github.io/en/posts/ntk/This post introduces Neural Tangent Kernel (NTK) theory, which underpins research focused on the dynamics of training.Tue, 09 Jan 2024 00:00:00 GMThttps://speed1313.github.io/en/posts/inshi2023/https://speed1313.github.io/en/posts/inshi2023/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.Thu, 23 Nov 2023 00:00:00 GMThttps://speed1313.github.io/en/posts/optimal-transport/https://speed1313.github.io/en/posts/optimal-transport/This post explains the basic ideas of optimal transport and some of its applications.Mon, 06 Nov 2023 00:00:00 GMThttps://speed1313.github.io/en/posts/os_zisaku/https://speed1313.github.io/en/posts/os_zisaku/In this post, I describe how building my own OS changed the way I see operating systems.Thu, 17 Aug 2023 00:00:00 GMThttps://speed1313.github.io/en/posts/gomicollector/https://speed1313.github.io/en/posts/gomicollector/This post explains the mark & sweep method, one of the algorithms used for garbage collection.Mon, 23 Jan 2023 00:00:00 GMThttps://speed1313.github.io/en/posts/cs_world/https://speed1313.github.io/en/posts/cs_world/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.Sun, 18 Dec 2022 00:00:00 GMThttps://speed1313.github.io/en/posts/sec-camp2022-memory/https://speed1313.github.io/en/posts/sec-camp2022-memory/I recently took part in Security Camp 2022, Track B (Web Security).Fri, 12 Aug 2022 00:00:00 GMT