Train a 64M-parameter language model from scratch in about two hours, with every step written in plain PyTorch.
git clone https://github.com/jingyaogong/minimindMiniMind walks you through building a small language model from zero. Every core algorithm, from pretraining to RLHF to tool calling, is implemented without high-level framework shortcuts.
Clone the repo, install the requirements and follow the README training steps.
Stanford NLP · Tool / CLI
Stanford's framework for programming, not prompting, language models, with optimizers that tune your prompts for you.
Prefect · Tool / CLI
The fast, Pythonic way to build MCP servers and clients, now a full MCP application framework from Prefect.
Promptfoo · Tool / CLI
Test and compare prompts, agents and RAG pipelines, and red-team them for security problems, from the command line.