As you surely find out working with chatgpt, for some longer answer chatgpt asking Paid Account users to click on ‘Continue generating’ button. In the dev environment, this is unlikely… any idea how to avoid that, if possible ?
“Continue” is because the ChatGPT maximum token generation limit has been hit for the response, the equivalent of API’s “max_tokens” parameter.
“Continue” is a feature unexposed in the API for developers, letting the model continue right where it left off without sending new messages.
The length cannot be changed and is set per-model selection by OpenAI.
I don’t care why, I care how.
I may have a solution… I would need some time to test it.
Instead of clicking the “Continue generating” button, you can achieve the same effect by sending a message with “continue” via the API.
When you click the “Continue generating” button in ChatGPT, it might be sending a “continue” message in the background.
Each model has a token limit for completions. There is one possible solution, control the number of words the LLM outputs using prompt engineering. 4 characters approximately equals to 1 token. You could do the math. Or mimic, “continue generating” button using chat completions. You could pass in the generated completions and related context to continue completions. Hope this helps. Cheers!
I want to avoid to click - that’s the idea !
Many times I don’t pay attention, focusing on another screen, and found after a while that chatgpt doesn’t finished the job but pause it, by asking me if to Continue. Simply annoying.
I don’t want to replace the click on Continue with typing a word - what’s the benefit ?!
Then you could omit that and write a function which automatically adjust the token count by removing unnecessary stuff and passing only the latest completions and related context. Cheers!
How about creating a script that automatically inputs “continue” and presses the Enter key?
I think there’s a Python module for that. ChatGPT might be able to help you with how to make it, I’m not sure though.
Tell me more about your idea, pls.
And how Python will read DOM? Seems complicated. Thanks.