AI Dataset2023
A 15,000+ English instruction–response corpus for fine-tuning and evaluating LLM instruction-following behavior. Contains human-authored prompts and answers across categories (closed/open QA, summarization, extraction, classification, brainstorming) and is released under CC BY-SA 3.0.
AI Dataset2023
Contains short, small-vocabulary stories synthetically generated by GPT-3.5 and GPT-4 for training and evaluating compact language models. Includes multiple splits, a GPT-4-only V2 subset, and archive files with prompts and metadata for reproducible experiments.
MLOps2023
Framework for unit-testing, evaluating and benchmarking LLM systems with ready-made metrics (G‑Eval, hallucination, task completion), support for local judge models and synthetic datasets, plus CI-friendly integrations for LangChain/OpenAI/Anthropic.
AI Others2023
Offline, privacy-first English grammar checker implemented in Rust that lints documents in milliseconds and uses a fraction of LanguageTool's memory. Can run as a local binary, embed via WebAssembly, and offers integrations for editors and plugins.
AI Dataset2023
Provides a 200k-example filtered conversational dataset derived from UltraChat for supervised fine‑tuning and generation‑ranking of chat models. Stored in parquet with four splits and used as part of Zephyr‑7B‑β training data.
AI Train2023
Collection of runnable model implementations — LLaMA, Mistral, Stable Diffusion, Whisper, CLIP, plus LoRA fine-tuning — ported to the MLX array framework so they run natively on Apple silicon's unified memory rather than CUDA.
AI Infra2023
Traces how Transformer LLMs route information from input to output, attributing each block's effect to individual attention heads and feed-forward neurons. Click any edge to see what a head promotes or suppresses in vocabulary space.
Re-derives LLM scaling laws, tracing prior disagreements to how compute budget was modeled, then trains 7B and 67B models on 2T tokens. The 67B model beats LLaMA-2 70B on code, math, and reasoning; its chat variant tops GPT-3.5 on open-ended evals.
Reworks the MoE layer to push each expert toward a narrow specialty: split experts into many finer ones and activate more per token, plus reserve a few always-on shared experts for common knowledge. A 2B model matches GShard 2.9B; at 16B it rivals LLaMA2 7B on ~40% of the compute.
A family of open code models (1.3B-33B) trained from scratch on 2T tokens of project-level code, using a 16K-window fill-in-the-blank objective. Beats Codex and GPT-3.5 on code benchmarks and ships under a license permitting commercial use.
AI Train2024
Runs GPT-4o-class vision, speech, and full-duplex audio-video conversation on a 9B model small enough to deploy on phones and tablets. The 4.5 release scores 77.6 on OpenCompass and adds real-time bilingual voice with voice cloning.
Reaches 51.7% on the competition-level MATH benchmark with a 7B model and no tools or voting, rivaling Gemini-Ultra and GPT-4. Built on a 120B-token math corpus mined from Common Crawl, and introduces GRPO, a memory-efficient PPO variant for reasoning.