[PYTHON] Add Sentiment Classification Generation Support

[ryanweiler92]

Add Sentiment Classification Generation Support

Overview

Implement sentiment classification generation class and related infrastructure to support both binary and 3-class sentiment analysis using CPU-friendly HuggingFace models. Use CPU friendly HF models such as distilbert-base-uncased-finetuned-sst-2-english and twitter-roberta-base-sentiment for development. Any new classes should follow the style of the existing classes.

Technical Requirements

1. Model Configuration

• Add new model type "sentiment" in model configuration
• Support for both binary and 3-class sentiment models:
  • Default binary: "distilbert-base-uncased-finetuned-sst-2-english"
  • Default multi-class: "twitter-roberta-base-sentiment"

2. Create SentimentGen Class

Create new file gaas/sentiment_gen/sentiment_gen.py:

• Implement SentimentGen class following existing pattern from similar classes
• Support both single text and batch inference
• Include proper error handling and logging

3. Update Main Application

• Add SentimentGen to model type initialization in main.py

elif model_type == "sentiment":
    app.state.gaas = SentimentGen(model_manager=model_manager)

4. Request/Response Models

• Create new Pydantic models in pydantic_models/

5. Integration Requirements

• Support responses through /api/generate endpoint

6. Update documentation

[pasupathimuniyappan]

• Started understanding the remote-client-server repo by going through the code flow and README document.
• Able to correlate this task with the existing code.

[pasupathimuniyappan]

Task Understanding - Implementation Steps

1. Create new branch and checkout

2. Model Configuration - [server/model_utils/model_config.py]

   • Add new model type "sentiment" in model configuration
   • elif model_type == "sentiment":
     return SentimentGenRequest(**request)

3. Model Usage - Support for both binary and 3-class sentiment models:

   • Have to read/explore about this below models, like usage and required things to include.
   • For binary: "distilbert-base-uncased-finetuned-sst-2-english"
   • For multi-class: "twitter-roberta-base-sentiment"

4. Create SentimentGen Class

   • Create new file - [gaas/sentiment_gen/sentiment_gen.py]
   • Implement SentimentGen class following existing pattern from similar classes
     -> create init - constructor
     -> create required functions to process the single and batch inference
     -> like load models, predict binary and multi-class classification using point 3
     -> Add error handling and logging

5. Add Request validation in Pydantic models - [server/pydantic_models/request_model.py]

   • Create class for SentimentGenRequest
   • Add the paramaters that we are required for this request.

6. Create API endpoint - [server/router/sentiment_gen_route.py]

   • create router function with below things,
     -> receive and validate the request using pydnatic models
     -> Create and queue the jobs using queue manager by calling approraite sentiment_gen.py load model function and predict function to get back the response
     -> convert the output in to proper response format(JSON)
     -> return as api response

7. Update Main Application - [server/main.py]

   • Add SentimentGen to model type initialization in main.py
   • elif model_type == "sentiment":
     app.state.gaas = SentimentGen(model_manager=model_manager)
   • app.include_router(sentiment_gen_route, prefix="/api")

8. Raise PR to master branch by adding Ryan as reviewer

9. Once approved, update the GitHub Repo documentation for remote-client-server repo

[pasupathimuniyappan]

Working on implementing the steps that I mentioned in previous comment.

[pasupathimuniyappan]

Update:

• Added the required codes and models for sentiment generation in the existing code.
• Formatting the code using Black
• Working on reviewing, testing and enhancing the code.

[pasupathimuniyappan]

Update:

• Completed the formatting, reviewing and testing.
• Raised the PR as well -> Integrated the sentiment generation model #28

[pasupathimuniyappan]

Update:

• Updated the code as per PR comments.
• Updated the PR as well.