Failed to Read Response about llm-workflow-engine HOT 5 CLOSED

The browser backend was removed in the lastest release: https://github.com/llm-workflow-engine/llm-workflow-engine/releases/tag/v0.15.0

You can see a brief explanation of why here: https://github.com/llm-workflow-engine/llm-workflow-engine/blob/main/CHATGPT_WRAPPER.md

I would be very surprised if you could not get as high quality answers from the API as you are from the ChatGPT app -- with the API you have full control over the system message, temperature, top_p, top_k, etc. I encourage you to play around with those!

FYI, this was the issue that really killed the browser backend for good: #311 -- despite me providing ample information in that issue, not a single person stepped forward to assist further with overcoming it, and I wasn't interested in it being a one-man show... ;)

from llm-workflow-engine.

from llm-workflow-engine.

Thanks for the quick response! The individual completions from the API are just as good as anything generated within ChatGPT. However the conversations within the site maintain context for much longer.

I gotta tell ya, this makes no sense to me. I've routinely had very long conversations via the API, and it's perfectly capable of maintaining the context of the conversation -- I recently had a GPT-4 conversation that was just over 8,000 tokens, and the limit is 8192. That worked perfectly.

If you're using GPT-4 models, then #311 is the issue.

But, for me, that ship has sailed. Perhaps somebody will fork the last 0.14.x release and revive the browser backend in another project.

from llm-workflow-engine.

from llm-workflow-engine.

If you're going to work w/ the API, you really should be aware of the model sizes and capabilities: https://platform.openai.com/docs/models/overview

And be aware of GPT-4 availability: https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4

from llm-workflow-engine.