Multimodal: <img x.jpg> will only detect filename, and ignore HTML #54
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Hey @BabyCNM... thanks for submitting this and it looks useful. Would you be able to comment with some short examples of how images and audio are currently included, and then updated examples which work with your code? I'm happy to test it out :) |
|
Here is an example where current implementation would fail but the edited version will work. prompt = """Read the screenshot image and the website's source code. Then, answer the user's question.
User Question: is the button below or above the image?
Screenshot: <img C:/User/xyz/Desktop/screenshot_3.jpg>
--- HTML Code ---
<!DOCTYPE html>
<html lang="en">
<img src="website/relative/path/300.jpg" alt="Placeholder Image">
<button onclick="alert('Button clicked!')">Click Me</button>
</body>
</html>
"""Note there are two locations we have "<img" tag. However, the first one should be interpreted as an image to be sent for GPT-4o, and the other one (which is embedded in HTML code) should just be code rather than image. |
odoochain
pushed a commit
to odoochain/autogen
that referenced
this pull request
Nov 10, 2024
odoochain
pushed a commit
to odoochain/autogen
that referenced
this pull request
Nov 10, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Why are these changes needed?
The autogen tag parsing system uses HTML-like tags to allow users to input images and audio directly from text. However, this system may mistakenly interpret actual HTML content (such as website source code) as multimodal components for GPT-4o and other VLMs, which is undesirable.
Fortunately, autogen’s tag format differs from HTML. In autogen, file paths do not require quotation marks. To improve parsing accuracy, we’ve introduced a strict_filepath_match parameter for the multimodal utilities. When enabled (True), it ensures that only simple tag contents—without spaces or quotes—are matched, making it especially useful for detecting filenames and ignoring complex HTML syntax. This parameter is turned on (True) for parsing multimodal agents' messages.
Note: This is a custom tagging convention, which could be confusion for some users. Please share any recommendations regarding the current design. Further simplification of the message component is planned for future updates.
Related issue number
Checks