From b4ca5636bc6e1d029809da9fccfc3dd69c01f53b Mon Sep 17 00:00:00 2001 From: Jake Poznanski Date: Mon, 18 Nov 2024 04:54:12 +0000 Subject: [PATCH] Decent set of todos for monday --- pdelfin/beakerpipeline.py | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) diff --git a/pdelfin/beakerpipeline.py b/pdelfin/beakerpipeline.py index 5bcfcc8..1e88c48 100644 --- a/pdelfin/beakerpipeline.py +++ b/pdelfin/beakerpipeline.py @@ -804,5 +804,11 @@ if __name__ == "__main__": asyncio.run(main()) # TODO - # Possible future addon, in beaker, discover other nodes on this same job - # Send them a message when you take a work item off the queue + # - Refactor the work queue into its own file so it's reusable and generic, and it makes temporary work files (prevent issue where if a work item is done, then it stalls because queue was just emptied) + # - Fix the queue release mechanism so that it just does a timeout, based on zero queue size only, so you don't block things + # - Add the page rotation check and mechanism + # - Sglang commit a fix for the context length issue + # - Get a solid benchmark on the stream vs non stream approach + # - sglang error on s3://ai2-s2-pdfs/73ee/35e7ed5c2fb113ceba652284aaa51db7c2fc.pdf-2 + # - Client error on attempt 0 for s3://ai2-s2-pdfs/e13c/9e03ce463ba53bfb15b26dbfd55c0bbc5568.pdf-1: 400, message='Bad Request', + # - Fix loading of the model checkpoints, it's so flakey now, maybe use datasets \ No newline at end of file