Spaces:

Duplicated from  trl-lib/stack-llama

bigcode

/

bigcode-playground

like 405

Runtime error

Hi,
I'm wondering if make sense to fine tune StarCoder on my own codebase to try to obtain better and more contextual response from the model.

A question that I'd like to ask is for example: "Create a Python integration module between mySystem1 and mySystem2 that allow all customer entities to be synced between the two systems"
Where:

• mySystem1 and mySystem2 are two custom application my team built and I own all the code bases
• "customer entities" must be translated in variable names based on the above codebases by the LLM

The only way to reach this goal is to fine tune a model like StarCoder? if yes, how can I prepare my dataset to train it? if not, are there other ways to do it?

Cheers,
Alexio