EmbeddedLLM / Phi-3-small-128k-instruct-onnx

Phi-3-small-128k-instruct ONNX

This repository hosts the optimized versions of microsoft/Phi-3-small-128k-instruct to accelerate inference with DirectML and ONNX Runtime. The Phi-3-Small-128K-Instruct is a state-of-the-art, lightweight open model developed by Microsoft, featuring 7B parameters.

Key Features:

• Parameter Count: 7B
• Tokenizer: Utilizes the tiktoken tokenizer for improved multilingual tokenization, with a vocabulary size of 100,352 tokens.
• Context Length: Default context length of 128k tokens.

Attention Mechanism:

• Implements grouped-query attention to minimize KV cache footprint, with 4 queries sharing 1 key.
• Uses alternative layers of dense attention and a novel blocksparse attention to further optimize on KV cache savings while maintaining long context retrieval performance.
• Multilingual Capability: Includes an additional 10% of multilingual data to enhance its performance across different languages.

ONNX Models

Here are some of the optimized configurations we have added:

• ONNX model for int4 DirectML: ONNX model for AMD, Intel, and NVIDIA GPUs on Windows, quantized to int4 using AWQ.
• ONNX model for int4 CPU and Mobile: ONNX model for CPU and mobile using int4 quantization via RTN. There are two versions uploaded to balance latency vs. accuracy. Acc=1 is targeted at improved accuracy, while Acc=4 is for improved performance. For mobile devices, we recommend using the model with acc-level-4.

Usage

Installation and Setup

To use the Phi-3-small-128k-instruct ONNX model on Windows with DirectML, follow these steps:

1. Create and activate a Conda environment:

conda create -n onnx python=3.10
conda activate onnx

1. Install Git LFS:

winget install -e --id GitHub.GitLFS

1. Install Hugging Face CLI:

pip install huggingface-hub[cli]

1. Download the model:

huggingface-cli download EmbeddedLLM/Phi-3-small-128k-instruct-onnx --include="onnx/directml/*" --local-dir .\Phi-3-small-128k-instruct

1. Install necessary Python packages:

pip install numpy==1.26.4
pip install onnxruntime-directml
pip install --pre onnxruntime-genai-directml

1. Install Visual Studio 2015 runtime:

conda install conda-forge::vs2015_runtime

1. Download the example script:

Invoke-WebRequest -Uri "https://raw.githubusercontent.com/microsoft/onnxruntime-genai/main/examples/python/phi3-qa.py" -OutFile "phi3-qa.py"

1. Run the example script:

python phi3-qa.py -m .\Phi-3-small-128k-instruct

Hardware Requirements

Minimum Configuration:

• Windows: DirectX 12-capable GPU (AMD/Nvidia/Intel)
• CPU: x86_64 / ARM64

Tested Configurations:

• GPU: AMD Ryzen 8000 Series iGPU (DirectML)
• CPU: AMD Ryzen CPU

Hardware Supported

The model has been tested on:

• GPU SKU: RTX 4090 (DirectML)

Minimum Configuration Required:

• Windows: DirectX 12-capable GPU and a minimum of 10GB of combined RAM

Model Description

• Developed by: Microsoft
• Model type: ONNX
• Language(s) (NLP): Python, C, C++
• License: MIT
• Model Description: This is a conversion of the Phi-3 Small 128K Instruct model for ONNX Runtime inference.

Additional Details

• Phi-3 Small, Medium, and Vision Blog
• Phi-3 Model Blog Link
• Phi-3 Model Card
• Phi-3 Technical Report
• Phi-3 on Azure AI Studio

License

The model is licensed under the MIT license .

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft’s Trademark & Brand Guidelines . Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party’s policies.