Andreas/LibreChat
1
0
Fork 0
mirror of https://github.com/danny-avila/LibreChat.git synced 2025-12-17 08:50:15 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 7

🖼️ fix: correct image extraction (#3538)

Browse source 
* fix: regex issue extracting text with images in markdown

* fix: update addImages function to include only the first observed image path in the response message

* ci: tests for addImages function: correct image extraction

* fix(GoogleClient): linting

---------

Co-authored-by: Dongwoo Jeong <dongwoo.jeong@lge.com>
Co-authored-by: Dongwoo Jeong <dongwoo@duck.com>
This commit is contained in:
Danny Avila 2024-08-04 20:53:11 -04:00 • committed by GitHub
parent 389b2a6cab
commit 5baa95bd9a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 86 additions and 26 deletions
Download patch file Download diff file Expand all files Collapse all files

6
api/app/clients/GoogleClient.js
View file

@ -879,7 +879,8 @@ class GoogleClient extends BaseClient {
return [ return [
{ {
category: 'HARM_CATEGORY_SEXUALLY_EXPLICIT', category: 'HARM_CATEGORY_SEXUALLY_EXPLICIT',
threshold: process.env.GOOGLE_SAFETY_SEXUALLY_EXPLICIT || 'HARM_BLOCK_THRESHOLD_UNSPECIFIED', threshold:
process.env.GOOGLE_SAFETY_SEXUALLY_EXPLICIT || 'HARM_BLOCK_THRESHOLD_UNSPECIFIED',
}, },
{ {
category: 'HARM_CATEGORY_HATE_SPEECH', category: 'HARM_CATEGORY_HATE_SPEECH',
@ -891,7 +892,8 @@ class GoogleClient extends BaseClient {
}, },
{ {
category: 'HARM_CATEGORY_DANGEROUS_CONTENT', category: 'HARM_CATEGORY_DANGEROUS_CONTENT',
threshold: process.env.GOOGLE_SAFETY_DANGEROUS_CONTENT || 'HARM_BLOCK_THRESHOLD_UNSPECIFIED', threshold:
process.env.GOOGLE_SAFETY_DANGEROUS_CONTENT || 'HARM_BLOCK_THRESHOLD_UNSPECIFIED',
}, },
]; ];
} }

6
api/app/clients/output_parsers/addImages.js
View file

@ -60,10 +60,10 @@ function addImages(intermediateSteps, responseMessage) {
if (!observation || !observation.includes('![')) { if (!observation || !observation.includes('![')) {
return; return;
} }
const observedImagePath = observation.match(/!\[.*\]\([^)]*\)/g); const observedImagePath = observation.match(/!\[[^(]*\]\([^)]*\)/g);
if (observedImagePath && !responseMessage.text.includes(observedImagePath[0])) { if (observedImagePath && !responseMessage.text.includes(observedImagePath[0])) {
responseMessage.text += '\n' + observation; responseMessage.text += '\n' + observedImagePath[0];
logger.debug('[addImages] added image from intermediateSteps:', observation); logger.debug('[addImages] added image from intermediateSteps:', observedImagePath[0]);
} }
}); });
} }

58
api/app/clients/output_parsers/addImages.spec.js
View file

@ -81,4 +81,62 @@ describe('addImages', () => {
addImages(intermediateSteps, responseMessage); addImages(intermediateSteps, responseMessage);
expect(responseMessage.text).toBe(`${originalText}\n${imageMarkdown}`); expect(responseMessage.text).toBe(`${originalText}\n${imageMarkdown}`);
}); });
it('should extract only image markdowns when there is text between them', () => {
const markdownWithTextBetweenImages = `
![image1](/images/image1.png)
Some text between images that should not be included.
![image2](/images/image2.png)
More text that should be ignored.
![image3](/images/image3.png)
`;
intermediateSteps.push({ observation: markdownWithTextBetweenImages });
addImages(intermediateSteps, responseMessage);
expect(responseMessage.text).toBe('\n![image1](/images/image1.png)');
});
it('should only return the first image when multiple images are present', () => {
const markdownWithMultipleImages = `
![image1](/images/image1.png)
![image2](/images/image2.png)
![image3](/images/image3.png)
`;
intermediateSteps.push({ observation: markdownWithMultipleImages });
addImages(intermediateSteps, responseMessage);
expect(responseMessage.text).toBe('\n![image1](/images/image1.png)');
});
it('should not include any text or metadata surrounding the image markdown', () => {
const markdownWithMetadata = `
Title: Test Document
Author: John Doe
![image1](/images/image1.png)
Some content after the image.
Vector values: [0.1, 0.2, 0.3]
`;
intermediateSteps.push({ observation: markdownWithMetadata });
addImages(intermediateSteps, responseMessage);
expect(responseMessage.text).toBe('\n![image1](/images/image1.png)');
});
it('should handle complex markdown with multiple images and only return the first one', () => {
const complexMarkdown = `
# Document Title
## Section 1
Here's some text with an embedded image:
![image1](/images/image1.png)
## Section 2
More text here...
![image2](/images/image2.png)
### Subsection
Even more content
![image3](/images/image3.png)
`;
intermediateSteps.push({ observation: complexMarkdown });
addImages(intermediateSteps, responseMessage);
expect(responseMessage.text).toBe('\n![image1](/images/image1.png)');
});
}); });