Cosmosage: A specialized AI assistant for cosmology
Cosmosage is an AI assistant specialized in cosmology. It has been trained on thousands of cosmology papers and textbooks, starting from a large-language model base and subsequently fine-tuned on cosmology-specific data and synthetic Q&A pairs. This specialized training enables it to answer domain-specific questions more accurately than general-purpose models.
We invite the community to try it out and share feedback, report bugs, or suggest features by leaving comments on the Hugging Face model page: https://huggingface.co/AstroMLab/AstroSage-70B
I assisted Tijmen de Haan with the deployment of this service on Jetstream 2.
Related posts
- Cosmosage Unshelver - Technical details on the deployment controller.
- Deploy a 70B LLM to Jetstream - Guide on deploying the underlying model infrastructure.
- Deploy a ChatGPT-like LLM on Jetstream with llama.cpp - A broader tutorial on running GGUF models on Jetstream, adapted from the work on Cosmosage.
- Timing the Unshelving of a Jetstream 70B LLM Instance - Measurements on how fast the service can come online from a shelved state.