The exact LLM used in the experiment mentioned in this post was upstage-llama-2-70b-instruct-v2.ggmlv3.q2_K. Grist was configured to use it via llama-cpp-python and <a href="https://github.com/gristlabs/grist-core#ai-formula-assistant-related-variables-all-optional">https://github.com/gristlabs/grist-core#ai-formula-assistant...</a>