feat/regex-replace-guardrail #1312
Conversation
|
DescriptionSummary By MatterAI
🔄 What ChangedThis PR introduces comprehensive support for multi-modal content, specifically image data, within the Google and Google Vertex AI chat completion providers. Key changes include adding an 🔍 Impact of the ChangeThis feature significantly enhances the capabilities of the Google and Google Vertex AI providers by enabling them to process and generate multi-modal content, primarily images. It allows for richer, more dynamic AI interactions where users can provide images as part of their prompts and potentially receive image-based responses, aligning the gateway with advanced multi-modal AI model functionalities. This expands the range of use cases and improves the overall user experience for applications leveraging these providers. 📁 Total Files Changed
🧪 Test AddedN/A 🔒Security VulnerabilitiesNo new security vulnerabilities are directly introduced by these changes, assuming the base64 encoded image data ( MotivationTo enable multi-modal capabilities (image input/output) for Google and Google Vertex AI chat completion providers, enhancing their functionality. Type of Change
How Has This Been Tested?
Screenshots (if applicable)N/A Checklist
Related Issues. Tip Quality Recommendations
Tanka Poem ♫
Sequence DiagramsequenceDiagram
participant User as User
participant Gateway as Portkey Gateway
participant GoogleAPI as Google/Vertex AI API
User->>+Gateway: POST /chat/completions (text, image_url, modalities)
Note over Gateway: Parse request body
Gateway->>Gateway: transformGenerationConfig(params)
Note over Gateway: Map modalities to responseModalities
Gateway->>Gateway: Process content_blocks
alt Image Data Present (inlineData)
Gateway->>GoogleAPI: POST /v1/models/gemini:generateContent
Note over GoogleAPI: Request with generationConfig, content parts (text, inlineData)
GoogleAPI-->>-Gateway: 200 OK (response with content parts)
Gateway->>Gateway: Parse API response
alt Response has inlineData
Gateway->>Gateway: Convert inlineData to image_url content_block
else Response has text/functionCall
Gateway->>Gateway: Process text/functionCall content_block
end
Gateway-->>-User: 200 OK (multi-modal response)
else Only Text Content
Gateway->>GoogleAPI: POST /v1/models/gemini:generateContent
Note over GoogleAPI: Request with generationConfig, content parts (text)
GoogleAPI-->>-Gateway: 200 OK (response with content parts)
Gateway->>Gateway: Parse API response
Gateway-->>-User: 200 OK (text response)
end
|
![matter-code-review[bot]](https://avatars.githubusercontent.com/in/474549?s=60&v=4){kind=small}
| ) => { | ||
| let error = null; | ||
| let verdict = true; | ||
| let data: any = null; |
There was a problem hiding this comment.
🐛 Bug Fix
Issue: Using any type weakens type safety and can lead to runtime errors
Fix: Use proper typing for the data object
Impact: Prevents potential runtime type errors and improves code maintainability
| let data: any = null; | |
| let data: { regexPattern?: string; verdict?: boolean; explanation?: string } | null = null; |
| if (!regexPattern) { | ||
| throw new Error('Missing regex pattern'); | ||
| } | ||
| if (!content) { | ||
| throw new Error('Missing text to match'); | ||
| } | ||
|
|
||
| const regex = new RegExp(regexPattern, 'g'); |
There was a problem hiding this comment.
⚡️ Performance Improvement
Issue: Creating new RegExp instance inside the handler for every execution is inefficient
Fix: Move regex creation after validation but before the loop to avoid recreation
Impact: Reduces object allocation overhead and improves performance for multiple text processing
| if (!regexPattern) { | |
| throw new Error('Missing regex pattern'); | |
| } | |
| if (!content) { | |
| throw new Error('Missing text to match'); | |
| } | |
| const regex = new RegExp(regexPattern, 'g'); | |
| if (!regexPattern) { | |
| throw new Error('Missing regex pattern'); | |
| } | |
| if (!content) { | |
| throw new Error('Missing text to match'); | |
| } | |
| const regex = new RegExp(regexPattern, 'g'); |
| const matches = text.match(regex); | ||
| if (matches && matches.length > 0) { | ||
| hasMatches = true; | ||
| } |
There was a problem hiding this comment.
🐛 Bug Fix
Issue: The match() method already returns null when no matches are found, making the length check redundant
Fix: Simplify the condition to just check for matches existence
Impact: Cleaner code and slightly better performance by avoiding unnecessary array length check
| const matches = text.match(regex); | |
| if (matches && matches.length > 0) { | |
| hasMatches = true; | |
| } | |
| const matches = text.match(regex); | |
| if (matches) { | |
| hasMatches = true; | |
| } |
| explanation: transformed | ||
| ? `Pattern '${regexPattern}' matched and was replaced with '${redactText}'` | ||
| : `The regex pattern '${regexPattern}' did not match any text.`, |
There was a problem hiding this comment.
🛠️ Code Refactor
Issue: The explanation message construction could be more readable and maintainable
Fix: Use template literals consistently and improve readability
Impact: Better code maintainability and consistent string formatting
| explanation: transformed | |
| ? `Pattern '${regexPattern}' matched and was replaced with '${redactText}'` | |
| : `The regex pattern '${regexPattern}' did not match any text.`, | |
| explanation: transformed | |
| ? `Pattern '${regexPattern}' matched and was replaced with '${redactText}'` | |
| : `The regex pattern '${regexPattern}' did not match any text.`, |
|
Code adds modalities support and inline data handling for Google Vertex AI. Implementation looks solid with proper type definitions and consistent patterns across files. |
2 participants
Description
This PR adds a new guardrail regex replace to portkey's default gurardrail
Motivation
the ability to replace text while running a regex guardrail on both input and output
Type of Change
How Has This Been Tested?
Checklist
Related Issues
Fixes and Cloaes #1291