File size: 588 Bytes
85c6f62
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
---
license: mit
pipeline_tag: text-generation
tags:
 - ONNX
 - DML
 - ONNXRuntime
 - phi3
 - nlp
 - conversational
 - custom_code
---

# Phi-3 Mini-128K-Instruct ONNX model for onnxruntime-web
This is the same models as the [official phi3 onnx model](https://huggingface.co/microsoft/Phi-3-mini-128k-instruct-onnx) with a few changes to make it work for onnxruntime-web:

1. the model is fp16 with int4 block quantization for weights
2. the 'logits' output is fp32 
3. the model uses MHA instead of GQA
4. onnx and external data file need to stay below 2GB to be cacheable in chromium