Using Large Action Models (LAMs) to automate workflow.

Title: Understanding Salesforce's Large Action Models (LAMs): Beyond Text Generation

Introduction:

Salesforce has recently introduced Large Action Models (LAMs), representing a significant advancement beyond traditional Large Language Models (LLMs). While LLMs excel at generating text and media, LAMs take AI capabilities a step further by enabling autonomous execution of complex tasks.

What are Large Action Models?

LAMs are designed to actively engage in business workflows, moving beyond mere text generation to perform actions like:

- Sending emails

- Managing CRM systems

- Making real-time decisions

- Integrating seamlessly with various tools, data, and systems

- Planning, executing, and adapting to user commands

These capabilities make LAMs particularly valuable for industries such as marketing, customer service, and beyond.

Technical Implementation:

The article demonstrates a practical implementation using the following steps:

1. Environment Setup:

- Utilizing L4 GPU

- Mounting to a specific drive

- Installing required libraries (Transformers, datasets, tokenizers, Flask)

2. Model Loading:

- Implementation of a 7-billion parameter model

- Loading both model and tokenizer using Transformers

3. Function Definitions:

The system uses OpenAI-format function definitions, including:

- Task instructions

- Format instructions adhering to JSON structure

- Specific function parameters and requirements

Example Functions:

1. Weather Function:

- Required parameters: location (mandatory)

- Optional parameters: unit (Celsius/Fahrenheit)

2. Search Function:

- Required parameters: query string

Practical Implementation:

The demonstration includes:

1. Flask REST API Creation:

- Development of proxy business processes

- Implementation of three main routes:

  a. Default health check route

  b. Customer information route (with ID parameter)

  c. Email sending route

2. API Integration:

- Creation of custom API specifications

- Implementation of actual function calls

- Response handling and parameter mapping

Results and Testing:

The system successfully:

- Interpreted natural language queries

- Mapped them to appropriate function calls

- Generated correct JSON responses

- Executed actual API calls with proper parameters

Future Implications:

While similar frameworks like Autogen and Crew exist, Salesforce's LAMs represent a focused approach to business process automation. The models are available on Hugging Face, making them accessible for testing and implementation.

Conclusion:

Large Action Models represent a promising development in AI, particularly for business applications. Their ability to not just understand but execute complex tasks makes them valuable tools for process automation and business workflow optimization.

This represents a significant step forward in making AI more actionable and practical for business applications, moving beyond simple text generation to actual task execution and process automation.

Link - Watch the video that this article is based on.