Status update on fine-tuning LLMs.
Tools to curate data and run economic analyses. I started by wrapping the GDP API from the Bureau of Economic Analysis and the inflation data from the Bureau of Labor Statistics.
A simple integration that lets you run Metaflow tasks with Ollama models as sidecars.
Built a PDF processing and chat app using Svelte for frontend and routing, FastAPI to serve the model, and Metaflow to run data ingestion workflows. The app is a simple chat interface that lets you upload a PDF and talk with an LLM that can reference specific pages of the document.
AWS Trainium is a new chip from AWS. It is a challenge, but not as hard as I expected to get it working with Metaflow. To run a model on Trainium, you need to use the AWS Neuron SDK, which uses the XLA backend for PyTorch.

emattia.com © 2025