The Part I video walked through embeddings and search with cosine similarity using the new v2 text-embedding-ada-002 model to provide in context learning from a dataset of markown files to an OpenAI completion call.
If haven't watched that video yet I recommend starting there:
• OpenAI v2 Embeddings +...
This video is follow-up that walks through some cleanup of the code, and some improvements to the rate limit handling code.
Code from this video can be found here:
github.com/Ops...
UPDATE:
-----------------
At some points when referring to the rate limit I use "per second" instead of "per minute". In all cases I am intending to say per minute.
So:
3,000 Requests Per Minute = OpenAI current limit for paid accounts older than 48 hours
60 Requests Per Minute = OpenAI current limit for paid accounts less than 48 hours old
20 Requests Per Minute = OpenAI Trial account limit
400-500 Requests Per Minute = How many embeddings requests I was able to send and get responses to when using the apply() on a Pandas dataframe.
00:00 Intro
00:27 Review error from last video
6:11 New rate limit code
9:48 53 million tokens correction
Негізгі бет OpenAI v2 Embeddings - Part 2 (text-embedding-ada-002)
Пікірлер