Model Overview

Description:

Model Developer: Microsoft

Phi-3.5-mini-Instruct INT4 ONNX

The NVIDIA Phi-3.5-mini-Instruct INT4 ONNX model is the quantized version of the Microsoft Phi-3.5-mini-Instruct model which has 3.8B parameters and is a dense decoder-only Transformer model using the same tokenizer as Phi-3 Mini. It supports 128K context length, therefore the model is capable of several long context tasks including long document/meeting summarization, long document QA, long document information retrieval. For more information, please check here. The NVIDIA Phi-3.5-mini-Instruct INT4 ONNX model is quantized with TensorRT Model Optimizer.

This model is ready for commercial and research use case.

Steps followed to generate this quantized model:

• Download Microsoft Phi-3.5-mini-Instruct model in Pytorch bfloat16 format from HuggingFace.

• Convert PyTorch model to ONNX FP16 using onnxruntime-genai model builder.

• Quantize Phi-3.5-mini-Instruct ONNX FP16 model to Phi-3.5-mini-Instruct ONNX INT4 AWQ model using TensorRT Model Optimizer – Windows.

Third-Party Community Consideration

This model is not owned or developed by NVIDIA. This model has been developed and built to a third-party’s requirements for this application and use case; see link to the non-NVIDIA Phi3.5-Mini-Model Card.

License/Terms of Use:

Governing Terms: Use of this model is governed by the NVIDIA Open Model License Agreement. Additional Information: MIT License.

Reference:

MIT License

Phi3.5-mini-Instruct Model Card

Readme

Model Architecture:

Phi-3.5-mini has 3.8B parameters and is a dense decoder-only Transformer model using the same tokenizer as Phi-3 Mini.

Architecture Type: Transformers

Network Architecture: Phi3

Input:

• Input Type: Text. It is best suited for prompts using chat format.

• Input Format: String

• Input Parameters: Sequence (1D)

• Other Properties Related to Input: Supports Arabic, Chinese, Czech, Danish, Dutch, English, Finnish, French, German, Hebrew, Hungarian, Italian, Japanese, Korean, Norwegian, Polish, Portuguese, Russian, Spanish, Swedish, Thai, Turkish, Ukrainian

Output:

• Output Type(s): Text

• Output Format: String

• Output Parameters: Sequence (1D)

Software Integration:

Supported Hardware Microarchitecture Compatibility: NVIDIA Ampere and newer GPUs. 6GB or higher GPU video memory are recommended. Higher VRAM may be required for larger context length use cases.

Supported Operating System(s): Windows

Model Version(s): v1.0

Training, Testing, and Evaluation Datasets:

Refer to Phi3.5-mini-Instruct Model Card for the details.

Training Dataset: cnn_dailymail used for calibration.

• Link: cnn_dailymail

• Data Collection Method by dataset: [Automated]

• Labeling Method by dataset: [Unknown]

Evaluation Dataset: MMLU

• Link:MMLU.

• Data Collection Method by dataset: [Unknown]

• Labeling Method by dataset: [Not Applicable]

Evaluation Results:

Accuracy Scores: MMLU (5 shots):

With GenAI ORT->DML backend, on a desktop RTX 4090 GPU system.

“overall accuracy” = 65.51

Test configuration:

• GPU: RTX 4090, RTX 3090.  

• Windows 11: 23H2

• NVIDIA Graphics driver: R565 or higher

Inference:

Inference Backend: Onnxruntime-GenAI-DirectML

(Note: please refer to Readme.txt for the detailed instructions.)

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.

Please report security vulnerabilities or NVIDIA AI Concerns here.