search

huangd1999 / AgentCoder

This Repo is the official implementation of AgentCoder and AgentCoder+.
237 stars 50 forks source link

Use local model #11

Open Shuyangg opened 3 weeks ago

Shuyangg commented 3 weeks ago

Hi, thanks for your work. I have a question when I try your codes. How can I apply this 3 agents work on my local model with local problems?

huangd1999 commented 3 weeks ago

Hi，I think the local model means that you want use open source LLMs such as Llama3 families, to employ these open source LLMs in AgentCoder, we need rewrite the code in Programer and Test Designer to apply the inference template (e.g., model.generate function)

then, its a good question for us to consider how to apply AgentCoder in local problem as not all problems are in JSON file with specific keys. To do that, currently we can remove the concurrecy function so that we can inference by task rank. Then, as only task descriptions are needed in our inference time. So that you can create a dictionary [ { “prompt “: task description}] in Programer file and save the model output dictionary into a json file. The json file can be used for other two agents.