ls articles --sort created

Articles

20 entries indexed

  1. OpenAI API-Compatible Access Without Additional API Billing via CodexA Codex-authenticated OpenAI API-compatible server for Responses, Chat Completions, and image generation. Within the Codex subscription scope, it can be used without additional API usage billing.
  2. Japanese Full-Text Search in SQLite and DuckDB with VaporettoWith the rise of various agents, embedded databases such as SQLite and DuckDB have become increasingly interesting because they run without a server and persist as local files.
  3. Building a Machine Learning PC with Two RTX 5090 GPUsI like training small Transformer models, usually around 100M parameters or less, and I built a custom PC with two RTX 5090 GPUs to improve training speed and learn more about multi-GPU setups.
  4. Looking Back on 2025Our child was born. People often say that the birth of a child is the biggest change in life, and now that I am living through it, I think that is exactly right.
  5. OpenProvence: A Model for Removing Irrelevant Sentences Before Passing Text to an LLMI released OpenProvence, a project for pruning irrelevant sentences from retrieved text before passing it to an LLM, with open source code and model weights.
  6. Evaluating the Japanese Performance of Embedding Gemma 300M with JMTEBGoogle recently released the embedding model google/embeddinggemma-300m, so I benchmarked its Japanese performance with JMTEB v1.
  7. JFWIR: A Large Japanese Information Retrieval Dataset Built from Japanese FineWebI released JFWIR, Japanese FineWeb Information Retrieval, a large dataset of about 64 million document-query pairs for Japanese information retrieval.
  8. Evaluating the Japanese Performance of Qwen3 Embedding with JMTEBThe open-weight multilingual Qwen3 Embedding and reranker models were released, so I measured Qwen3-Embedding-0.6B on JMTEB.
  9. Releasing Small, Fast, and Practical Japanese Rerankers: tiny, xsmall, small, and base v2I released small Japanese reranker models that run at practical speed on CPU and Apple silicon while keeping competitive retrieval quality.
  10. query-crafter-japanese: A Model for Generating Queries for Information RetrievalI released query-crafter-japanese, a small model that can generate retrieval queries from documents quickly and without output license restrictions.
  11. FineWeb2 Edu Japanese: A High-Quality Educational Japanese DatasetI published FineWeb2 Edu Japanese, a high-quality educational Japanese dataset.
  12. Releasing a Japanese StaticEmbedding Model for Practical 100x Faster Text EmbeddingsI released static-embedding-japanese, a Japanese and English StaticEmbedding model that can create practical text embeddings very quickly on CPU.
  13. Looking Back on 2024The year is coming to an end, so here is a short look back at 2024.
  14. Releasing Japanese SPLADE v2, a Strong Retrieval Model for Texts Under 512 TokensI released japanese-splade-v2, an improved Japanese SPLADE model with strong JMTEB retrieval performance for documents up to 512 tokens.
  15. Releasing Japanese BERT RetroMAE Models and Evaluating Them on Downstream Retrieval TasksI pretrained Japanese BERT models with RetroMAE, released the models, and evaluated their effect on downstream retrieval tasks with JMTEB.
  16. How to Build a SPLADE Model: Japanese SPLADE Technical ReportThis report describes the implementation, training, and evaluation of a Japanese SPLADE sparse retrieval model.
  17. Releasing a High-Performance Japanese SPLADE Sparse Retrieval ModelI created and released a Japanese SPLADE sparse vector model for text retrieval, with competitive results on retrieval and reranking tasks.
  18. Running Japanese Tokenizer Models with text-embeddings-inferenceHugging Face's text-embeddings-inference is a fast production inference server, but many Japanese models cannot be used directly because they do not provide tokenizer.json.
  19. Releasing High-Performance Japanese Rerankers, and What Rerankers AreI built a family of reranker models trained specifically for Japanese, from small models to larger ones, and evaluated them on Japanese reranking tasks.
  20. Technical Report on Building Japanese RerankersThis is a technical report on building Japanese reranker, or CrossEncoder, models.