From ed802fe2256b96318c5dc83361eb881a805e4572 Mon Sep 17 00:00:00 2001 From: rchan Date: Fri, 8 Sep 2023 18:18:14 +0100 Subject: [PATCH] add meeting notes for 08/09 --- meeting_notes/Reginald 08_09_23.md | 26 ++++++++++++++++++++++++++ 1 file changed, 26 insertions(+) create mode 100644 meeting_notes/Reginald 08_09_23.md diff --git a/meeting_notes/Reginald 08_09_23.md b/meeting_notes/Reginald 08_09_23.md new file mode 100644 index 00000000..b49aecb0 --- /dev/null +++ b/meeting_notes/Reginald 08_09_23.md @@ -0,0 +1,26 @@ +# Reginald 08/09/23 + +## Notes +- Ryan and Rosie almost finishing a PR contribution to llama-index for the following issue: https://github.com/jerryjliu/llama_index/issues/7596 + - ReAct engine seems to be a bit broken after this due to the default system prompt + - Potentially can make another issue and PR to fix +- Other potential llama-index contributions + - Some llama-cpp chat engine example notebooks - there's not many (if at all) that exist in the repo and they're very welcome to more examples + - Ryan to try fix old issue posted during Hackweek (https://github.com/jerryjliu/llama_index/issues/6465) +- Rosie been working on getting GitHub issues and files from within our Hut23 + - Will need to think about what is the personal token that we should use + - Maybe look at if there's an instituitional token that exists + - Will need to upload this to Azure as a secret +- Idea: look into logging how good the answers are and save into a database + - For future, we can maybe use this with a RLHF fine-tuning of our LLM + +## Actions +- Ryan & Rosie: look at putting this azure + - Figure out how to have multiple chat instances at the same time + - Can start with running the bot locally +- Rosie to think about using the reader for pure markdown files so no need to process them to csv + - Alternatively, to figure out how the csv files were created from pure markdown files + - Maybe it was from Andy's scripts to pull data from wiki and handbook +- Ryan to see if we can update the OpenAI code and compare with Llama2 +- Ryan & Rosie: Try to get a quantized Llama2-70b running (for Tomas) +- Merge Rosie's current PRs at some point next week