🐛 Fix bug that arises when using negative max_tokens #937

asim-shrestha · 2023-07-03T00:46:29Z

About

We can request a negative max token value when the completion tokens are that significant

This results in a strange response from OpenAI:

-2473 is less than the minimum of 1 - 'max_tokens'

This PR will ensure we have a minimum value of 1 max_tokens. This will result in the following response:

This model's maximum context length is 4097 tokens. However, your messages resulted in 6478 tokens. Please reduce the length of the messages.

Also added some tests for this method and moved it into the token_service

vercel · 2023-07-03T00:46:34Z

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Updated (UTC)
agent-gpt	✅ Ready (Inspect)	Visit Preview	Jul 3, 2023 0:46am
docs	✅ Ready (Inspect)	Visit Preview	Jul 3, 2023 0:46am

awtkns · 2023-07-03T00:49:12Z

platform/reworkd_platform/services/tokenizer/token_service.py

+from reworkd_platform.web.api.agent.model_settings import WrappedChatOpenAI
+
+
+class TokenService:


awtkns · 2023-07-03T00:49:43Z

Love the tests ❤️

ergomake · 2023-07-03T00:52:31Z

Hi 👋

Here's a preview environment 🚀

Container	Source	URL
next	Dockerfile	https://next-reworkd-agentgpt-937.env.ergomake.link
docs	Dockerfile	https://docs-reworkd-agentgpt-937.env.ergomake.link
platform	Dockerfile	https://platform-reworkd-agentgpt-937.env.ergomake.link
db	Dockerfile	[not exposed - internal service]
weaviate	semitechnologies/weaviate:1.19.6	https://weaviate-reworkd-agentgpt-937.env.ergomake.link

Questions? Comments? Suggestions? Join Discord.

Click here to disable Ergomake.

🐛 Fix bug that arises when using negative max_tokens

00d2c08

awtkns approved these changes Jul 3, 2023

View reviewed changes

asim-shrestha merged commit ee22203 into main Jul 3, 2023

		from reworkd_platform.web.api.agent.model_settings import WrappedChatOpenAI


		class TokenService: