Spaces:
Runtime error
Runtime error
ashish rai
commited on
Commit
·
0d4914a
1
Parent(s):
19da0ee
added script for onnx sent clf
Browse files- sentiment_onnx_classify.py +67 -0
sentiment_onnx_classify.py
ADDED
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import onnxruntime as ort
|
2 |
+
import torch
|
3 |
+
from transformers import AutoTokenizer
|
4 |
+
import numpy as np
|
5 |
+
|
6 |
+
tokenizer=AutoTokenizer.from_pretrained("sentiment_classifier/")
|
7 |
+
|
8 |
+
#create onnx & onnx_int_8 sessions
|
9 |
+
session=ort.InferenceSession("sent_clf_onnx/sentiment_classifier_onnx.onnx")
|
10 |
+
session_int8=ort.InferenceSession("sent_clf_onnx/sentiment_classifier_onnx_int8.onnx")
|
11 |
+
|
12 |
+
|
13 |
+
|
14 |
+
def classify_sentiment_onnx(texts,_model=session,_tokenizer=tokenizer):
|
15 |
+
"""
|
16 |
+
user will pass texts separated by comma
|
17 |
+
"""
|
18 |
+
try:
|
19 |
+
texts=texts.split(',')
|
20 |
+
except:
|
21 |
+
pass
|
22 |
+
|
23 |
+
_inputs = _tokenizer(texts, padding=True, truncation=True,
|
24 |
+
return_tensors="np")
|
25 |
+
|
26 |
+
input_feed={
|
27 |
+
"input_ids":np.array(_inputs['input_ids']),
|
28 |
+
"attention_mask":np.array((_inputs['attention_mask']))
|
29 |
+
}
|
30 |
+
|
31 |
+
output = _model.run(input_feed=input_feed, output_names=['output_0'])[0]
|
32 |
+
|
33 |
+
output=np.argmax(output,axis=1)
|
34 |
+
output = ['Positive' if i == 1 else 'Negative' for i in output]
|
35 |
+
return output
|
36 |
+
|
37 |
+
def classify_sentiment_onnx_quant(texts, _model=session_int8, _tokenizer=tokenizer):
|
38 |
+
"""
|
39 |
+
user will pass texts separated by comma
|
40 |
+
"""
|
41 |
+
try:
|
42 |
+
texts=texts.split(',')
|
43 |
+
except:
|
44 |
+
pass
|
45 |
+
|
46 |
+
_inputs = _tokenizer(texts, padding=True, truncation=True,
|
47 |
+
return_tensors="np")
|
48 |
+
|
49 |
+
|
50 |
+
input_feed={
|
51 |
+
"input_ids":np.array(_inputs['input_ids']),
|
52 |
+
"attention_mask":np.array((_inputs['attention_mask']))
|
53 |
+
}
|
54 |
+
|
55 |
+
output = _model.run(input_feed=input_feed, output_names=['output_0'])[0]
|
56 |
+
|
57 |
+
output=np.argmax(output,axis=1)
|
58 |
+
output = ['Positive' if i == 1 else 'Negative' for i in output]
|
59 |
+
|
60 |
+
return output
|
61 |
+
|
62 |
+
|
63 |
+
|
64 |
+
|
65 |
+
|
66 |
+
|
67 |
+
|