The current state of the repo cannot be run by anyone who wants to try out auto-llama. This commit contains the changes necessary to make that happen with a fresh install.
Changes
All references to the openAI package have been removed. This included removing the dalle image generation and reworking the token counter.
There are a few locations where runtime errors were caused by type errors. These runtime errors were the result of changing the llm library from openAI to llama where library functions didn't exactly match up with the same return or parameter types. Issues here are llama_cpp max_tokens requires an Int and create_embedding returned a dict instead of a list of floats.
The llama model_path was hardcoded to the dev's file path. This should be part of the env file (well, there's a lot of other "better" directions that could be taken here, but this will work best for now).
PR Quality Checklist
[X] My pull request is atomic and focuses on a single change.
While it is multiple smaller changes, it is all required for the main change of "get this to actually run so more dev work can be done later"
[X] I have thoroughly tested my changes with multiple different prompts.
[X] I have considered potential risks and mitigations for my changes.
[X] I have documented my changes clearly and comprehensively.
[X] I have not snuck in any "extra" small tweaks changes
LoireLab
commented
1 year ago
Background
The current state of the repo cannot be run by anyone who wants to try out auto-llama. This commit contains the changes necessary to make that happen with a fresh install.
Changes
All references to the openAI package have been removed. This included removing the dalle image generation and reworking the token counter.
There are a few locations where runtime errors were caused by type errors. These runtime errors were the result of changing the llm library from openAI to llama where library functions didn't exactly match up with the same return or parameter types. Issues here are llama_cpp max_tokens requires an Int and create_embedding returned a dict instead of a list of floats.
The llama model_path was hardcoded to the dev's file path. This should be part of the env file (well, there's a lot of other "better" directions that could be taken here, but this will work best for now).
PR Quality Checklist