From 8e827b36f91b456862583473a8181f8b3223d460 Mon Sep 17 00:00:00 2001 From: rchan Date: Mon, 2 Oct 2023 14:20:05 +0100 Subject: [PATCH] =?UTF-8?q?=F0=9F=A7=91=E2=80=8D=F0=9F=92=BB=2002/10=20mee?= =?UTF-8?q?ting?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- meeting_notes/Reginald_10_02_23.md | 37 ++++++++++++++++++++++++++++++ 1 file changed, 37 insertions(+) create mode 100644 meeting_notes/Reginald_10_02_23.md diff --git a/meeting_notes/Reginald_10_02_23.md b/meeting_notes/Reginald_10_02_23.md new file mode 100644 index 00000000..5f1220e8 --- /dev/null +++ b/meeting_notes/Reginald_10_02_23.md @@ -0,0 +1,37 @@ +# Reginald 02/10/23 + +## Notes +- Rosie summary of what we've done in the FM + - Got Llama-2 model running (with `llama-cpp-python` to run quantised versions) + - Implemented chat engine which can remember chat history instead of a query engine + - Implemented queue - add stuff to a (asyncio) queue and process queries once at a time + - You get a clock emoji to say that it's been registered but not yet been processed - this gets changes to a llama when it is being processed + - Idea for future work to tell you the position in the queue if possible (using the various clock emojis as a count down) + - Using readers from `llama-hub` to obtain data from different sources + - Used to have all documents (from wikis, handbooks, turing website, a few Turing courses) inside the repo, but now able to remove the data - just need a `GITHUB_TOKEN` environment variable + - Can only access Hut23 (in `all_data` index) if you're invited to the repo + - Almost finished implementing API for machine hosting LLM and the machine hosting the bot + - Ryan to pick this up and try finish + - Made several PRs to `llama-index` and `llama-hub` repos + - Updated README + - Updated Dockerfiles in the repo +- Remaining tasks to do: + - Pulumi scripts + - Testing the splitting of the model and the bot (API) properly + - It's all been on a local machine at the moment + - Once we have that, we might be able to actually host the llama-cpp model properly and make it available for some light testing within REG + - Need to update the blob storage indexes + - Tech talk to update progress and maybe try to get people to join the project +- Future work + - Turing slack? + - Get feedback on team and try get people from REG to help out with specific engineering problems + - Improve chat history management - can get errors after filling up the context. There is a temprorary fix to clear history using a slack command + + - Evaluation and comparison of models + +## Actions +- Rosie to message Turing IT +- Ryan wrap up API stuff +- Get tech talk time in maybe end of November +- Get monthly date in the calendar for Reginald catch up and 22 days work +