File size: 2,049 Bytes
3353b9f
f2b1f40
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3353b9f
f2b1f40
 
 
3353b9f
f2b1f40
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
language:
- en
- ar
- zh
- fr
- de
- it
- ja
- ko
- nl
- pl
- pt
- es
- th
- tr
- ru
tags:
- feature-extraction
- onnx
- use
- text-embedding
- tensorflow-hub
license: apache-2.0
inference: false
widget:
- text: Thank goodness ONNX is available, it is lots faster!
---
### Universal Sentence Encoder Multilingual v3

ONNX version of [https://tfhub.dev/google/universal-sentence-encoder-multilingual/3](https://tfhub.dev/google/universal-sentence-encoder-multilingual/3)

The original TFHub version of the model is referenced in other models here E.g. [https://huggingface.co/vprelovac/universal-sentence-encoder-large-5](https://huggingface.co/vprelovac/universal-sentence-encoder-large-5) 

### Overview

See overview and license details at [https://tfhub.dev/google/universal-sentence-encoder-multilingual/3](https://tfhub.dev/google/universal-sentence-encoder-multilingual/3)

This model is a full precision version of the TFHub original, in ONNX format.

It uses the [ONNXRuntime Extensions](https://github.com/microsoft/onnxruntime-extensions) to embed the tokenizer within the ONNX model, so no seperate tokenizer is needed, and text is fed directly into the ONNX model.

Post-processing (E.g. pooling, normalization) is also implemented within the ONNX model, so no separate processing is necessary.

### How to use

```python
import onnxruntime as ort
from onnxruntime_extensions import get_library_path
from os import cpu_count

sentences = ["hello world"]

def load_onnx_model(model_filepath):
  _options = ort.SessionOptions()
  _options.inter_op_num_threads, _options.intra_op_num_threads = cpu_count(), cpu_count()
  _options.register_custom_ops_library(get_library_path())
  _providers = ["CPUExecutionProvider"]  # could use ort.get_available_providers()
  return ort.InferenceSession(path_or_bytes=model_filepath, sess_options=_options, providers=_providers)

model = load_onnx_model("filepath_for_model_dot_onnx")

model_outputs = model.run(output_names=["outputs"], input_feed={"inputs": sentences})[0]
print(model_outputs)
```