Skip to content

Conversation

gmontamat
Copy link
Contributor

@gmontamat gmontamat commented Aug 18, 2025

@carlrobertoh: this is the only feature missing in ProxyAI to finally ditch Github Copilot!

FIM Code completion is done in ProxyAI using gpt-3.5-turbo-instruct via the /v1/completions endpoint only which is now legacy. Many OpenAI-compatible APIs do not support this endpoint.

I added Chat-based FIM code completion and tested with gpt-4.1.

May conflict with: #1090, I could rebase if necessary.

Screenshot_20250818_130647

Sample GPT-4.1 chat completion:

Screenshot_20250821_114946

@carlrobertoh
Copy link
Owner

Looks good. However, I believe the UI should be made more clearer. I had to follow the exact configuration from your screenshots to quickly understand how to test this logic. I think this approach should be applied to every provider and model that isn't trained on FIM tokens.

@carlrobertoh carlrobertoh merged commit 550b0e8 into carlrobertoh:master Sep 3, 2025
2 checks passed
carlrobertoh added a commit that referenced this pull request Sep 3, 2025
* Implement FIM code completion with a chat models

* Fix chat-based FIM

* fix: build

---------

Co-authored-by: Carl-Robert Linnupuu <carlrobertoh@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants