File size: 2,105 Bytes
01b5826
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e7128e7
01b5826
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e8b5dfc
 
 
 
 
 
 
 
 
 
01b5826
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
---
license: mit
pipeline_tag: text-generation
tags:
 - ONNX
 - DML
 - ONNXRuntime
 - phi3
 - nlp
 - conversational
 - custom_code
inference: false
language:
- en
---
# EmbeddedLLM/Phi-3-mini-4k-instruct-062024-int4-directml

## Model Summary

This model is an ONNX-optimized version of [microsoft/Phi-3-mini-4k-instruct (June 2024)](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct), designed to provide accelerated inference on a variety of hardware using ONNX Runtime(CPU and DirectML).
DirectML is a high-performance, hardware-accelerated DirectX 12 library for machine learning, providing GPU acceleration for a wide range of supported hardware and drivers, including AMD, Intel, NVIDIA, and Qualcomm GPUs.

## 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.


### Hardware Requirements

**Minimum Configuration:**
- **Windows:** DirectX 12-capable GPU (AMD/Nvidia) 
- **CPU:** x86_64 / ARM64
**Tested Configurations:**
- **GPU:** AMD Ryzen 8000 Series iGPU (DirectML)
- **CPU:** AMD Ryzen CPU
  
## Model Description
- **Developed by:** Microsoft
- **Model type:** ONNX
- **Language(s) (NLP):** Python, C, C++
- **License:** Apache License Version 2.0
- **Model Description:** This model is a conversion of the Phi-3-mini-4k-instruct-062024 for ONNX Runtime inference, optimized for DirectML.

## Performance Metrics

<!-- These are the evaluation metrics being used, ideally with a description of why. -->
### DirectML
We measured the performance of DirectML on AMD Ryzen 9 7940HS /w Radeon 78

| Prompt Length | Generation Length | Average Throughput (tps) |
|---------------------------|-------------------|-----------------------------|
| 128 | 128  | - |
| 128 | 256  | - |
| 128 | 512 | - |
| 128 | 1024 | - |
| 256 | 128  | - |
| 256 | 256  | - |
| 256 | 512 | - |
| 256 | 1024 | - |
| 512 | 128  | - |
| 512 | 256  | - |
| 512 | 512 | - |
| 512 | 1024 | -  |
| 1024 | 128  | - |
| 1024 | 256  | - |
| 1024 | 512 | - |
| 1024 | 1024 | -  |