DOCSEARCHbyAlgolia
Submit Docs

NVIDIA ACE

repository·main·Indexed 43 days ago

https://github.com/nvidia/ace

A suite of generative AI technologies and microservices (NIMs) designed to power digital humans, interactive NPCs, and AI-driven customer service avatars. Includes tools for deploying RAG bots, Stock bots, and Colang 2.0 sample bots via Kubernetes and Helm, with support for multimodal interaction management and speech interfaces.

Tokens
35.8K
Snippets
108
Records
148
Agent score
55%

What's inside NVIDIA ACE

  2. Overview of ACE Agent Bot Web UI features

    main

    The ACE Agent Bot Web UI is a web application designed for interacting with ACE Agent bots. It supports three primary modes of interaction:

    • Text conversations: Standard text-based chat with the bot.
    • Speech conversations: Enables real-time voice interaction. Users can speak via microphone and hear audio responses from the bot. The UI provides real-time Automatic Speech Recognition (ASR) transcripts for both the user and the bot.
    • Emojis: For bots capable of expressing non-verbal signals (gestures and emotions), the UI translates these signals into emojis in both text and speech modes.

  3. Overview of Food Ordering Bot features

    main

    The Food Ordering Bot is a virtual assistant built using Colang designed to assist users with food orders. Its core functionalities include:

    • Listing menu items: Displaying available food options.
    • Managing order cart: Adding, removing, and replacing items in the user's cart.
    • Listing bill amount: Calculating and showing the total cost.
    • Event handling: Demonstrating the use of entry/exit events within the agent workflow.

  4. Overview of Audio2Face Microservice resources

    main

    The Audio2Face (A2F) Microservice repository provides the necessary components to deploy and interact with the Audio2Face service. Key resources include:

    • gRPC Definitions: Located in the proto/ folder, these contain the service definitions and installation instructions for the gRPC interfaces.
    • Deployment: The quick-start/ folder contains a docker-compose configuration for deploying both the Audio2Face and A2F Controller containers.
    • Example Applications: The scripts/ folder contains sample applications demonstrating how to use the A2F Microservice.
    • Test Data: The example_audio/ folder contains audio files for testing the service.

  5. Overview of Gaming NPC Bots

    main

    The Gaming Non-Playing Character (NPC) sample bots demonstrate how to build LLM-driven Natural Language Understanding (NLU) and Natural Language Generation (NLG) capabilities for game characters. These bots allow you to provide game-stage specific context to the Chat Engine at runtime to dynamically change NPC behavior.

    Two sample bots are provided:

    • Jin: Personality and backstory defined in samples/jin/bot_config.yaml.
    • Elara: Personality and backstory defined in samples/elara/bot_config.yaml.

    Both bots can be deployed in a single Chat Engine instance and use the mixtral-8x7b-instruct model from the NVIDIA API Catalog.

  6. Overview of ACE Agent

    main

    ACE Agent is a collection of microservices designed to build scalable, customizable, LLM-driven Conversational AI Agents. It provides a complete workflow for deploying virtual agents capable of multi-turn and multi-user contextual conversations.

    Key capabilities include:

    • LLM Integration: Out-of-the-box support for Large Language Models with hooks to connect custom models.
    • On-premise Deployment: Supports on-premise deployment of ACE models and custom models via NVIDIA NIM for GPU-accelerated serving.
    • Customization via Colang: Allows developers to define and customize bot behavior using Colang.
    • Framework Integration: Supports integrating agents built with LangChain or similar frameworks into the ACE Agent pipeline.
    • RAG Support: Easy integration with Retrieval Augmented Generation (RAG) workflows for knowledge-based agents.
    • Low Latency: Optimized using NVIDIA TensorRT, NVIDIA Triton Inference Server, and an optimized chat controller.

  7. Overview of NVIDIA Unified Cloud Services (UCS) Tools

    main

    NVIDIA Unified Cloud Services (UCS) Tools is a low-code framework designed for building cloud-native, real-time, and multimodal AI applications. It uses a Microservices Architecture approach, allowing developers to combine specialized microservices into complex applications.

    Key features include:

    • Low-code design tools: For developing microservices and applications.
    • Domain-specific microservices: Each microservice operates within a bounded domain context (e.g., Vision AI, Conversational AI, Animation AI & Rendering, Data Analytics) and can be independently deployed, managed, and scaled.
    • Distributed execution: Applications can run across multiple devices, from the cloud to edge and embedded platforms.
    • Extensibility: Domain experts can create new microservices using NVIDIA SDKs that follow UCS-compatible specifications and design guidelines.

    For detailed implementation guides and examples, refer to the official UCS Tools documentation.

  8. Spanish Bot capabilities and features

    main

    The Spanish Bot, built using Colang, is designed to handle real-time weather data, temporal information, and open-domain questions in Spanish.

    Supported features include:

    • Weather Data: Forecasts, temperature, wind speed, humidity, precipitation, and weather conditions (e.g., Sunny or Cloudy) for specific locations.
    • Temporal Information: Current date and time.
    • General Knowledge: Open domain Q&A in Spanish.

  9. Capabilities of the Spanish NMT Bot

    main

    The Spanish NMT bot is designed to handle Spanish-language queries by translating them to English for processing and back to Spanish for responses. It supports the following functionalities:

    • Weather Information: Forecasts, temperature, wind speed, humidity, precipitation, and weather conditions (e.g., Sunny or Cloudy) for specific locations.
    • Temporal Information: Current date and time.
    • General Knowledge: Open domain Question and Answering (QnA).

  12. What is Colang 2.0?

    main

    Colang 2.0 is the latest iteration of the Colang language and runtime. It builds upon the guardrailing and conversational interaction management of Colang 1.0 but adds core support for multimodality.

    Key features include:

    • Parallel actions
    • Hierarchical flows
    • UMIM actions and events
    • Multimodal interaction management between users and one or more bots.

    In ACE Agent, Colang 2.0 is accessed via an asynchronous, event-based interface.