Getting gibberish responses from 2.7.5 build.

[Steve Gehrman]

Bug Report

I compile the backend and use it in my own app. I was using 2.7.1 which works fine, but just tried to update to 2.7.5 and it fails.

Any idea why? I'm sending NULL to fake_reply. Is that OK? what is that for exactly?

Steps to Reproduce

compile on linux, use in an app.

Your Environment

• GPT4All version: 2.7.5
• Operating System: Linux
• Chat model used (if applicable): all of them

[Steve Gehrman]

It doesn't work on 2.7.2, so something changed in that build.

[Jared Van Bortel]

PR #2014 was merged between 2.7.1 and 2.7.2, so if you're using a binding it probably needs to be updated to reflect the new layout of struct llmodel_prompt_context.

Otherwise, I'm going to need more information about how you are using the GPT4All backend.

[Steve Gehrman]

Thanks! adding the min_p seems to have fixed it!