// field guide
Why foundations matter
An axiom is a statement accepted as the starting point for reasoning. Modern AI has its own axioms: the transformer architecture, gradient descent at scale, the datasets models learn from, and the published research that tells us whether progress is real. Axiomi is a directory of the tools that embody those foundations. It is written for researchers, students and engineers who want to understand and work with models at the level of weights, code and data rather than through a chat window.
The open weight models section is the heart of the site. Families such as Llama, Qwen, Gemma, Mistral, DeepSeek, OLMo and gpt-oss can be downloaded, studied and fine tuned. They differ in licence, size, training transparency and intended use, and the gap between them and closed frontier systems has narrowed substantially. We note the GitHub repository for each where one exists, and the live trending panel shows which models are gaining attention on Hugging Face right now.
Below the models sit the frameworks that make them possible. PyTorch dominates research, JAX powers much of Google's work and large scale TPU training, and MLX and llama.cpp bring models to laptops. Training and fine tuning tools (DeepSpeed, Megatron-LM, TRL, PEFT, Unsloth and Axolotl) turn a base model into something useful. Interpretability libraries such as TransformerLens, nnsight and SAELens let you look inside the network and ask what it has actually learned, a question that grows more important as models become more capable.
Finally, foundations rest on data and on open research. The datasets section covers open corpora like Common Crawl, FineWeb and Dolma, the research platforms section gathers the preprint servers, model hubs, review sites and hosted notebooks where results are published and reproduced, and the interpretability section shows how researchers test what a model has actually learned. The research feed pulls the newest papers from arXiv and Hugging Face Daily Papers, so you can move from a tool to the work behind it in a single click. Every description on Axiomi is original, every link goes to the official source, and every number comes from a public source with its date.