Model Overview

Description:

USD Code is an OpenUSD Python code generation and knowledge answering model that helps developers to write OpenUSD code and answer OpenUSD knowledge questions.

The following NIM are used by USD Code:

• Llama-3.1-70b-instruct
• nv-embedqa-e5-v5

Llama-3.1-70b-instruct is used to drive the code generation and the agentic workflow, while NVIDIA Retrieval QA E5 Embedding is used for Retrieval Augmented Generation (RAG) to answer OpenUSD knowledge questions, USD code generation, and high-level Helper Function-based code generation. Helper Functions provide high-level abstractions leveraging the USD API, enabling developers to efficiently manage complex tasks such as creating, modifying, and querying USD scene.

This model is ready for commercial use.

Licenses:

If you download the software and materials as available from the NVIDIA AI product portfolio, use is governed by the NVIDIA Software License Agreement and the Product-Specific Terms for NVIDIA AI Products; except for the model which is governed by the NVIDIA AI Foundation Models Community License Agreement, and the RAG dataset which is governed by the terms of the NVIDIA Asset License.

ADDITIONAL INFORMATION: For Llama model, Llama 3.1 Community License Agreement, Built with Llama; for NV-EmbedQA-E5-v5: MIT license; for NV-EmbedQA-Mistral7B-v2: Apache 2.0 license, and Snowflake arctic-embed-l: Apache 2.0 license.

If you download the software and materials as available from the NVIDIA Omniverse portfolio, use is governed by the NVIDIA Software License Agreement and the Product-Specific Terms for NVIDIA Omniverse; except for the model which is governed by the NVIDIA AI Foundation Models Community License Agreement, and the RAG dataset which is governed by the terms of the NVIDIA Asset License.

ADDITIONAL INFORMATION: For Llama model, Llama 3.1 Community License Agreement, Built with Llama; for NV-EmbedQA-E5-v5: MIT license; for NV-EmbedQA-Mistral7B-v2: Apache 2.0 license, and Snowflake arctic-embed-l: Apache 2.0 license.

References:

• Llama-3.1 - https://ai.meta.com/blog/meta-llama-3-1/
• NVIDIA Retrieval QA E5 Embedding - https://build.nvidia.com/nvidia/nv-embedqa-e5-v5/
• OpenUSD - https://www.openusd.org/

Model Architecture:

• Llama-3.1-70b-instruct
  • Architecture Type: Transformer
  • Network Architecture: Llama-3.1
• NVIDIA Retrieval QA E5 Embedding
  • Architecture Type: Transformer
  • Network Architecture: Fine-tuned E5-Large-Unsupervised retriever

Input

Input Type(s): Text
Input Format(s): String
Input Parameter(s): One Dimentional (1D)
Other Properties Related to Input: Max context length of 128k tokens

Output

Output Type(s): Text (Code, Python)
Output Format: String
Output Parameter(s): One Dimentional (1D)
Other Properties Related to Output: Max output length of 128k tokens

Software Integration:

Runtime Engine(s):

• TensorRT

Supported Hardware Microarchitecture Compatibility:

• NVIDIA Ampere
• NVIDIA Hopper

[Preferred/Supported] Operating System(s):

• Linux

Model Version(s):

• Llama-3.1
  • llama-3.1-70b-instruct:1.3.0
• NVIDIA Retrieval QA E5 Embedding
  • nv-embedqa-e5-v5:1.1.0

Training Dataset:

• Llama-3.1-70b-instruct

  • Llama-3.1-70b-instruct were not trained by NVIDIA.
  • Link: https://build.nvidia.com/meta/llama-3_1-70b-instruct/modelcard
  • Data Collection Method: Hybrid: Human, Synthetic
  • Labeling Method: Unknown
  • Description: The 70B Llama 3.1 model is trained on a new mix of publicly available online data, supports multilingual text input and output (including code), has a 128k context length, 15T+ tokens, GQA enabled, and a knowledge cutoff of December 2023.

• NVIDIA Retrieval QA E5 Embedding

  • Link: https://build.nvidia.com/nvidia/nv-embedqa-e5-v5/modelcard
  • Data Collection Method: Unknown
  • Labeling Method: Unknown
  • Description: The model is trained on 400k samples from public datasets licensed for commercial use, focused on English (US) for information retrieval and question answering over text documents, with data collection and labeling methods unspecified.

Inference:

Engine:

• TensorRT

Test Hardware:

• A100
• H100

Ethical Considerations:

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
For more detailed information on ethical considerations for this model, please see the Model Card++ Explainability, Bias, Safety & Security, and Privacy Subcards. Please report security vulnerabilities or NVIDIA AI Concerns here.