Model linespan endpoint #601
Conversation
There was a problem hiding this comment.
@Free-Quarks , have you tested this locally with the other project deps? We're using pydantic v2 in this project. I've found that to be incompatible with langchain's prompt templates. To ensure skema isn't going to hit the same issue, can you please add a test that instantiates a prompt template?
|
Downgrading pydantic to a different major version might cause unexpected behavior. If you run into this issue in the test, maybe you can get around using the langchain prompt template by creating your own prompt string. |
Thanks for pointing this out! I am using the generic output parser from langchain, not their pydantic one, so hopefully it won't be an issue. I'm using pydantic just to simplify the final serialized json for FastAPI. But I will test it today. |
|
@myedibleenso So I was already running on pydantic 2.3, but I also updated to 2.4.1 and tested locally and it still works on that as well. So there shouldn't be an issue with pydantic v2! |
This is for an added endpoint that takes in a zip file containing a python script and returns the line span the dynamics should be under. It currently uses an Openai LLM for this, but we will replace that with a custom local model once it's ready. --------- Co-authored-by: Justin <[email protected]> 6009eb1
## Summary of Changes #601 introduced LLMs using `langchain` and the Open AI API. The `llms` extra was included in the extra `all`, but our docker image only installs `core` to avoid bloat: https://github.com/ml4ai/skema/blob/3af73f74ec127e0fb4c5abf4d3a8f9bd0eea3e71/Dockerfile.skema-py#L53 This caused an import error that triggered a deployment failure loop.
## Summary of Changes #601 introduced LLMs using `langchain` and the Open AI API. The `llms` extra was included in the extra `all`, but our docker image only installs `core` to avoid bloat: https://github.com/ml4ai/skema/blob/3af73f74ec127e0fb4c5abf4d3a8f9bd0eea3e71/Dockerfile.skema-py#L53 This caused an import error that triggered a deployment failure loop. be313e6
This is for an added endpoint that takes in a zip file containing a python script and returns the line span the dynamics should be under.
It currently uses an Openai LLM for this, but we will replace that with a custom local model once it's ready.