The NVIDIA Phi-4-multimodal-instruct FP8 model is the quantized version of Microsoft’s Phi-4-multimodal-instruct model, which is a multimodal foundation model that uses an optimized transformer architecture. For more information, please check here. The NVIDIA Phi-4-multimodal-instruct FP8 model is quantized with TensorRT Model Optimizer.

This model is ready for commercial/non-commercial use.

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 Non-NVIDIA (Phi-4-multimodal-instruct) Model Card.

License/Terms of Use:

Use of this model is governed by nvidia-open-model-license ADDITIONAL INFORMATION: MIT_License.

Deployment Geography:

Global, except in European Union

Use Case:

Developers looking to take off the shelf pre-quantized models for deployment in AI Agent systems, chatbots, RAG systems, and other AI-powered applications.

Release Date:

Huggingface 09/15/2025 via https://huggingface.co/nvidia/Phi-4-multimodal-instruct-FP8

Model Architecture:

Architecture Type: Transformers
Network Architecture: Phi4MMForCausalLM

**This model was developed based on Phi-4-multimodal-instruct ** Number of model parameters 5.6*10^9

Input:

Input Type(s): Text, image and speech
Input Format(s): String, Images (see properties), Soundfile
Input Parameters: One-Dimensional (1D), Two-Dimensional (2D), One-Dimensional (1D)
Other Properties Related to Input: Any common RGB/gray image format (e.g., (".jpg", ".jpeg", ".png", ".ppm", ".bmp", ".pgm", ".tif", ".tiff", ".webp")) can be supported. Any audio format that can be loaded by soundfile package should be supported. Context length up to 128K

Output:

Output Type(s): Text
Output Format: String
Output Parameters: 1D (One-Dimensional): Sequences
Other Properties Related to Output: N/A

Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA’s hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions.

Software Integration:

Supported Runtime Engine(s):

TensorRT-LLM

Supported Hardware Microarchitecture Compatibility:

NVIDIA Blackwell

Preferred Operating System(s):

Linux

Model Version(s):

The model is quantized with nvidia-modelopt v0.35.0

Post Training Quantization

This model was obtained by quantizing the weights and activations of Phi-4-multimodal-instruct to FP8 data type, ready for inference with TensorRT-LLM. Only the weights and activations of the linear operators within transformer blocks of the language model are quantized.

Training and Testing Datasets:

** Data Modality [Audio] [Image] [Text] ** Text Training Data Size [1 Billion to 10 Trillion Tokens] ** Audio Training Data Size [More than 1 Million Hours] ** Image Training Data Size [1 Billion to 10 Trillion image-text Tokens]

Calibration Dataset:

** Link: cnn_dailymail
** Data collection method: Automated.
** Labeling method: Automated.

Training Datasets:

** Data Collection Method by Dataset: Automated
** Labeling Method by Dataset: Human, Automated
** Properties: publicly available documents filtered for quality, selected high-quality educational data, and code

newly created synthetic, “textbook-like” data for the purpose of teaching math, coding, common sense reasoning, general knowledge of the world (e.g., science, daily activities, theory of mind, etc.)
high quality human labeled data in chat format
selected high-quality image-text interleave data
synthetic and publicly available image, multi-image, and video data
anonymized in-house speech-text pair data with strong/weak transcriptions
selected high-quality publicly available and anonymized in-house speech data with task-specific supervisions
selected synthetic speech data
synthetic vision-speech data

Testing Dataset:

** Data Collection Method by Dataset: Undisclosed
** Labeling Method by Dataset: Undisclosed
** Properties: Undisclosed

Inference:

Engine: TensorRT-LLM
Test Hardware: B200 coming soon
** Currently supported on DGX Spark

Usage

Deploy with TensorRT-LLM

To deploy the quantized checkpoint with TensorRT-LLM LLM API, follow the sample codes below:

LLM API sample usage:

from tensorrt_llm import LLM, SamplingParams


def main():

    prompts = [
        "Hello, my name is",
        "The president of the United States is",
        "The capital of France is",
        "The future of AI is",
    ]
    sampling_params = SamplingParams(temperature=0.8, top_p=0.95)

    llm = LLM(model="nvidia/Phi-4-multimodal-instruct-FP8", trust_remote_code=True)

    outputs = llm.generate(prompts, sampling_params)

    # Print the outputs.
    for output in outputs:
        prompt = output.prompt
        generated_text = output.outputs[0].text
        print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")


# The entry point of the program needs to be protected for spawning processes.
if __name__ == '__main__':
    main()

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 model quality, risk, security vulnerabilities or NVIDIA AI Concerns here.