[Firebase AI] Add Thought summaries #1317
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Summary of Changes
Hello @a-maurice, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!
This pull request introduces the capability to retrieve and expose the AI model's internal 'thought summaries' when generating content. This new feature allows developers to gain insight into the model's reasoning process, enhancing transparency and debugging capabilities for AI-powered applications. It includes updates to the content generation response structure and configuration options to enable this functionality.
Highlights
- New Feature: Thought Summaries: A new "ThoughtSummary" property has been added to "GenerateContentResponse" to provide a consolidated string of the model's internal thinking process, if available and enabled.
- Configuration for Thought Summaries: The "ThinkingConfig" within "GenerationConfig" now includes an "IncludeThoughts" option, allowing users to explicitly request thought summaries in the model's responses.
- Internal Content Part Identification: An "IsThought" property has been introduced to the "ModelContent.Part" interface and its implementing classes ("TextPart", "InlineDataPart", "FunctionCallPart") to differentiate between regular content and parts representing the model's thoughts.
- Behavioral Change in FunctionCalls Property: The "FunctionCalls" property in "GenerateContentResponse" now filters results to only include function calls that are explicitly marked as 'thoughts'. This changes its previous behavior of returning all function calls.
- Improved JSON Serialization/Deserialization: New "AddIfHasValue" extension methods have been added to "InternalHelpers.cs" to streamline the serialization of optional values to JSON dictionaries. Parsing logic for content parts has been updated to handle the new "IsThought" and "thoughtSignature" fields.
- Automated Testing: New automated tests ("TestIncludeThoughts", "InternalTestThoughtSummary") have been added to validate the correct functionality and parsing of thought summaries.
Using Gemini Code Assist
The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.
Invoking Gemini
You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.
| Feature | Command | Description |
|---|---|---|
| Code Review | /gemini review |
Performs a code review for the current pull request in its current state. |
| Pull Request Summary | /gemini summary |
Provides a summary of the current pull request in its current state. |
| Comment | @gemini-code-assist | Responds in comments when explicitly tagged, both in issue comments and review comments. |
| Help | /gemini help |
Displays a list of available commands. |
Customization
To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.
Limitations & Feedback
Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.
You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.
Footnotes
-
Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request adds support for receiving thought summaries from the model, which is a great feature for providing insight into the model's reasoning process. The implementation is mostly solid, with good additions to GenerationConfig and ModelContent to support this. However, I've identified two critical issues. First, the FunctionCalls property in GenerateContentResponse is incorrectly filtered to only return function calls that are thoughts, which is likely the opposite of the intended behavior. Second, in ModelContent, the InlineDataPart does not correctly deserialize the isThought property from JSON, which would cause it to be ignored. I've provided specific suggestions to address both of these issues.
| public static void AddIfHasValue<T>(this JsonDict jsonDict, string key, | ||
| T? value) where T : struct { | ||
| if (value.HasValue) { | ||
| jsonDict.Add(key, value.Value); | ||
| } | ||
| } | ||
|
|
||
| public static void AddIfHasValue<T>(this JsonDict jsonDict, string key, | ||
| T value) where T : class { | ||
| if (value != null) { | ||
| jsonDict.Add(key, value); | ||
| } | ||
| } |
There was a problem hiding this comment.
why we would need to functions, the only difference I can find is the ? after T in parameters
There was a problem hiding this comment.
By default, struct's are not-nullable, which is why I need the version with T?, and retrieve the underlying value with .Value. Classes can be null, and by default doing T? with a class will result in a warning. So doing two methods helps prevent that warning.
Description
Add logic for receiving thought summaries when generating content.
Testing
Running tests locally.
Type of Change
Place an
xthe applicable box: