Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[WIP] RolePlay prompts and behavioural improvements #11

Open
wants to merge 5 commits into
base: main
Choose a base branch
from

Conversation

kuzaxak
Copy link
Contributor

@kuzaxak kuzaxak commented Jul 7, 2024

Check commits for details. Overall the goal to prevent model to add commits like:

The system message for the AI reviewer has been refined and expanded. This update provides more specific guidance on review focus areas and exclusions. It should lead to more targeted and actionable feedback. Consider monitoring the quality of reviews after this change to ensure it meets your expectations.

Which aren't helpful.

kuzaxak added 4 commits July 7, 2024 19:24
This way we will make a clear separation between input and expected output for the model.
Based on my tests Clause models struggle to read complex rules without that and treat input chunks as examples of changes.

In addition to that lang prompt moved to a system one.

metada removed as it wan't used anywhere.
Commit message commonly answer on question why the change was made and will help in reducing pointless suggestions from LLM which already taken into account.

Overall prompt restrucutred according to Anthropic style guide (bulletpoints and heading via MD).
@kuzaxak kuzaxak force-pushed the commits-context branch from 52732d7 to 85ecad0 Compare July 7, 2024 16:25
@kuzaxak kuzaxak changed the title RolePlay prompts and behavioural improvements [WIP] RolePlay prompts and behavioural improvements Jul 7, 2024
```
    {
      "line_start": 103,
      "line_end": 104,
      "comment": "Similar to the previous comment, the formatting improvement here is good for consistency and readability."
    }
  ],
  "lgtm": true
```

Bedrock models respond with useless comments confirming that all good with lgtm set to `true`.

```
  "reviews": [
    {
      "line_start": 30,
      "line_end": 30,
      "comment": "The `image_tag_mutability` attribute is set to a boolean value (`true`), but it should be a string value of either 'MUTABLE' or 'IMMUTABLE'. Consider updating this line to:\n\n```diff\n-  image_tag_mutability = try(each.value[\"image_tag_mutability\"], true)\n+  image_tag_mutability = try(each.value[\"image_tag_mutability\"], \"MUTABLE\")\n```\n\nThis change ensures that the default value is 'MUTABLE', which aligns with AWS's default behavior and maintains backward compatibility as mentioned in the commit message."
    }
  ],
  "lgtm": false
}
```

In case of actual comment `lgtm` set to `false`
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant