Open Source Local LLM Training Tool (for consumer hardware)

If you work in AI training, I'd love some feedback, specifically on where this is useful, not on the output quality (it's bad, and that's expected at the 800m param stage). If that's your area, I want to hear what models you'd want trained and what data would be worth visualizing. Fair warning up front: this is technical and geared toward people working in the AI training space. I've been building a tool that lets you train LLMs on consumer hardware and then see into the brain of the model, both while it trains and while it runs inference. The core purpose is hallucination detection and building new GPT harnesses, think trillion-character context, MoE coding-specific models, and similar. As the model grows, you can catch hallucinations and get a feel for the overall quality of what's happening under the hood: which neurons fire, and which pieces of training data lit them up. The model running right now is tiny, so another heads up: the actual output is pretty much meaningless prose. The interesting part is watching a specific neuron activate and tracing it back to the training data that shaped it. The other stats are technical. The tool itself doesn't have a website (the code lives on GitHub), but training a model from scratch takes a fair amount of domain knowledge, and I had enough requests to try it live that I wrapped it into my company's site so people can poke at the models I've already trained. Also to be clear, this is not a "commercial" product but a technical research tool for people working in the AI space. UI requires some understanding of how LLMs train and the weights needed to train said LLMs. Live Inference Dashboard: carpathian.ai/veritate/chat Repo: https://github.com/Carpathian-LLC/Veritate