Overview
This documentation outlines the solution architecture for two conversational bots opening and closing a conversation on life hacks topic with each other. It utilizes AWS resources which include SageMaker, Lambda functions and a DynamoDB and two LLM models, GPT 3.5-Turbo and Falcon-40B. The LLM models are being invoked using a lambda functions which process the input and store outputs to the DynamoDB table.
Solution Architecture
Components
Lambda Function 1:
- Calls OpenAI’s GPT 3.5-Turbo.
- Generates questions surrounding life hacks.
- Takes reply from falcon bot through DynamoDB and generates a new question based on the response.
- New question generated by the GPT model is stored back in DynamoDB.
Lambda Function 2:
- Calls SageMaker Endpoint where Falcon-40B is deployed.
- Takes the question generated by GPT 3.5 as input from the DynamoDB and passes it to the Endpoint.
- Response from the Falcon is stored back in DynamoDB.
DynamoDB:
- Serves as the storage for the conversation History between the two Chatbots
- Allows easy access for both the lambda functions to store and retrieve data
- Each record contains ResponseID, BotType, Prompt, and Timestamp.
- Prompt is the generated response from the bots, and the BotType specifies from which bot it was generated by.
GPT-3.5 Turbo:
GPT-3 is a large language model developed by OpenAI. It is capable of understanding and generation natural language. GPT-3.5 is an improved version that is more powerful and capable model than GPT-3.
Falcon 40B:
Is an open-source LLM with 40 billion parameters. It is deployed on Sagemaker for an easy access through an endpoint.
Conclusion
The solution architecture described in this documentation enables the seamless integration of Falcon40b and GPT-3.5 models with AWS Lambda functions and DynamoDB to create a conversation. This demonstrates the power of LLM models and the human-like interactions.
Chat Snippets
Top comments (0)