How best_practices.md is being used?

[mattyan1726]

Hello, I am trying to understand how best_practices.md is being used in this document section
Is this relevant_best_practices here retrieving info from best_practices.md file and add to the LLM context? I am not seeing where best_practices.md is being used in the code.
Thank you!

[mattyan1726]

Also, I hope to understand the effect of adding instructions in the content of best_practices section in .pr_agent.toml vs extra_instructions in pr_review section in .pr_agent.toml. Thank you so much!

[mrT23]

best_practice file is a feature of pr-agent pro (there is a "💎" sign).
So you wont find the code using it in the open source

[mattyan1726]

@mrT23 thank you for the answer!
We subscribing to the pro version of pr-agent, and we are wondering if we contribute to the open source code (standard version), will that have an impact on the pro version? Or the pro version is a different forked project and contributing to the open source code will have no impact?

Also, I am observing review quality being different for e.g. java code vs IOS code. If I understand correctly, the default model was finetuned with 25k PR's, and I am wondering what these 25k PR's are composed of (distribution of PR's across programming languages). It also mentions that the pro version can configure it to use solely claude or gpt model here, are these models finetuned as well? What are the models we can experiment with in the config file other than 'claude-3-5-sonnet' and 'gpt-4o'.

Thanks again

[mrT23]

• (relevant) contribution to the open source will also be available for pro, in addition to being available in (regular) pr-agent

• PR-Agent Pro uses by default a combination of Claude-3.5-Sonnet and GPT-4o. For code suggestions, for example, Claude-3.5 is used. You can manually switch between the models by:
  https://qodo-merge-docs.qodo.ai/usage-guide/PR_agent_pro_models/#qodo-merge-pro-models
  For other tips regarding optimizing quality, see:
  https://qodo-merge-docs.qodo.ai/tools/improve/#extra-instructions-and-best-practices

• A fine-tuned model is used only for dedicated enterprise on-prem deployment, where companies prefer not to send their code to external LLMs.