AI Dataset2024
Hisami Suzuki, Satoru Katsumata +4LLM-jp (Center for Large Language Model Research and Development, National Institute of Informatics), National Institute of Informatics
Provides manually curated Japanese instruction pairs (questions and safe reference answers) for improving LLM output safety, covering broad harm categories and regionally sensitive cases. Includes English meta-tags and standard splits for benchmarking and fine-tuning.
AI Infra2025
Measures generative AI inference performance with token-level metrics (TTFT, inter-token latency), latency, and throughput under realistic traffic patterns. Provides a multiprocess engine, real-time TUI dashboard, extensible plugins, and integrations for telemetry and result uploads, aimed at inference benchmarking and capacity planning.
AI Agent2025
Evaluates and optimizes AI agents and language models in containerized environments, supporting large-scale parallel benchmarks and RL rollouts. Integrates with third‑party providers for thousands of parallel environments and serves as the official harness for Terminal‑Bench.
AI Dataset2025
Provides a reproducible, deduplicated corpus of text extracted from PDFs for LLM pretraining—about 3 trillion tokens from ~475 million documents in 1733 language-script pairs. Includes OCR and text extraction pipelines, per-page language IDs, MinHash deduplication, and is released under ODC‑By 1.0.
AI Dataset2026
Provides a physical reconstruction benchmark of OmniDocBench v1.5 by producing five real-world photographic variants (Scanning, Warping, Screen‑Photography, Illumination, Skew) for each of 1,355 pages, inheriting original ground-truth to enable controlled, scenario-wise evaluation of document parsing robustness.
MCP Server2026
Provides a local MCP server that returns precise, symbol-level code (functions, classes, imports) via tree-sitter parsing so AI agents send only the bytes they need—commonly cutting code-reading token usage 95%+ and enabling compact packed responses for further savings.
AI Agent2026
Runs a persistent, self-modifying AI agent locally with durable identity, memory, and versioned history across tasks. Provides native desktop and headless CLI runtimes, coordinated subagent swarms, configurable remote or local GGUF models, and reviewed self-evolution via Git.
AI Train2026
Provides a one-command CLI to fine-tune and post-train LLMs, with layer streaming that lets an 8B model be fine-tuned on a 4 GB laptop GPU. Auto-configures quantization, LoRA adapters, batching and evaluation gates, and supports export and serving workflows.
AI Dataset2026
Provides 6,000 runnable, operator-level PyTorch tasks for training and evaluating CUDA kernel generation models; each sample includes executable code, operator descriptors, and provenance tags, with execution-driven filtering to ensure reproducibility and contamination control.
AI Train2026
Runs transformer forward+backward training directly on Apple's Neural Engine by reverse-engineering private ANE APIs; includes per-layer ANE kernels, INT8 optimizations and benchmarks. Proof-of-concept only—relies on undocumented APIs, has low utilization and CPU fallbacks.
AI Dataset2026
Benchmarks LLM agents on realistic legal work by packaging lawyer-style assignments with client materials and expert, per-deliverable rubrics. Includes an execution harness to run, score, and compare agents across a large, evolving task set spanning multiple practice areas.
AI Agent2026
Monitors and detects risky behavior in enterprise AI agents via high-fidelity telemetry, security benchmarking, and a two-tier detector. Comprises ADR Sensor, ADR-Bench, and ADR Detector; deployed in production at Uber and validated on public benchmarks.