ACE & Agentic LLM Overview

Introduction to ACE LLM:

ACE LLM by Gupshup is a family of generative models to automate marketing, commerce & support related conversations powered by Open & Closed source foundational models. It has been fine-tuned on industry & function-specific data to deliver a delightful customer experience. Enterprises get a head-start in delivering ACE LLM-powered conversational experiences.

The ACE LLM has been built keeping the focus on compliance, accuracy & cost efficiency. AI guardrails implemented in the ACE family ensure ethical use, safety in responses, preventing harmful outputs or biased information from delivering the accurate results.

Recently, We have upgraded our LLM models in to continuously improve the performance and accuracy of our AI systems.

ACE Family Models (Till 11.0 release)Updated ACE Family Models (12.0 release)Use Case Served
ACE LLM Lite (Flan t5, Llama 2)ACE Lite (Llama-3.1-8b)Llama-3.1-8b: Intent & Entity Classification, FAQ Answering
ACE LLM Hybrid (Flan T5, Azure Open AI)ACE Pro (Gemma-2-2b, Azure OpenAI GPT-4o-mini)Gemma-2-2b: Intent & Entity Classification

GPT-4o-mini: FAQ Answering
ACE LLM Hybrid v2 (Mistral, Azure Open AI)ACE Advanced (Llama-3.1-8b, Azure OpenAI GPT-4o-mini)Llama-3.1-8b: Intent & Entity Classification

GPT-4o-mini: FAQ Answering
ACE LLM Classic (Azure Open AI)ACE Premium (Azure OpenAI GPT-4o-mini)GPT-4o-mini: Intent & Entity Classification, FAQ Answering

Introduction to Agentic LLM:

Gpt 4O & 4O-Mini are the core engines that drive the intelligence, response generation, and decision-making capabilities of your AI agents within the Gupshup AI Admin Module.

These models are used during both intent classification & skill execution and are configurable at both agent and skill level.

ModelDescriptionBest For
GPT-4o (AI Agent Premium)High-performance large language model from with strong reasoning & multi-modal capabilitiesComplex use cases, longer replies, premium accuracy
GPT-4o Mini (AI Agent Pro)Lightweight variant offering faster responses at a lower compute costUtility tasks, real-time replies, high-volume scenarios

What’s Next