Spaces:
Runtime error
Runtime error
Commit
·
1e516ec
1
Parent(s):
56e9c00
updated app.py with multiple models
Browse files
app.py
CHANGED
@@ -1,45 +1,69 @@
|
|
1 |
import gradio as gr
|
2 |
import torch
|
3 |
-
from transformers import AutoTokenizer, AutoModelForTokenClassification
|
4 |
-
|
5 |
-
tokenizer = AutoTokenizer.from_pretrained("ai4bharat/IndicNER")
|
6 |
-
|
7 |
-
model = AutoModelForTokenClassification.from_pretrained("ai4bharat/IndicNER")
|
8 |
-
|
9 |
-
|
10 |
-
def get_ner(sentence):
|
11 |
-
tok_sentence = tokenizer(sentence, return_tensors='pt')
|
12 |
|
|
|
|
|
13 |
with torch.no_grad():
|
14 |
-
logits = model(**
|
15 |
predicted_tokens_classes = [
|
16 |
-
|
|
|
17 |
|
18 |
predicted_labels = []
|
19 |
-
|
20 |
previous_token_id = 0
|
21 |
-
word_ids =
|
22 |
for word_index in range(len(word_ids)):
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
ner_output = []
|
32 |
-
for index in range(len(sentence.split(' '))):
|
33 |
-
ner_output.append(
|
34 |
-
(sentence.split(' ')[index], predicted_labels[index]))
|
35 |
return ner_output
|
36 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
37 |
|
38 |
-
iface = gr.Interface(
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
|
45 |
-
iface.launch(enable_queue=True)
|
|
|
1 |
import gradio as gr
|
2 |
import torch
|
3 |
+
from transformers import AutoTokenizer, AutoModelForTokenClassification,pipeline
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
|
5 |
+
def get_ner_bio(pipe,text):
|
6 |
+
tok_text = pipe.tokenizer(text, return_tensors='pt')
|
7 |
with torch.no_grad():
|
8 |
+
logits = pipe.model(**tok_text).logits.argmax(-1)
|
9 |
predicted_tokens_classes = [
|
10 |
+
pipe.model.config.id2label[t.item()] for t in logits[0]
|
11 |
+
]
|
12 |
|
13 |
predicted_labels = []
|
|
|
14 |
previous_token_id = 0
|
15 |
+
word_ids = tok_text.word_ids()
|
16 |
for word_index in range(len(word_ids)):
|
17 |
+
if not (word_ids[word_index] == None or word_ids[word_index] == previous_token_id):
|
18 |
+
predicted_labels.append(predicted_tokens_classes[word_index])
|
19 |
+
previous_token_id = word_ids[word_index]
|
20 |
+
|
21 |
+
ner_output = [
|
22 |
+
(word, label if label!="O" else None)
|
23 |
+
for word, label in zip(text.split(" "),predicted_labels)
|
24 |
+
]
|
|
|
|
|
|
|
|
|
25 |
return ner_output
|
26 |
|
27 |
+
def get_ner(pipe,text,aggregation_strategy="first"):
|
28 |
+
|
29 |
+
if aggregation_strategy == "bio_first":
|
30 |
+
return get_ner_bio(pipe,text)
|
31 |
+
else:
|
32 |
+
results = pipe(text,aggregation_strategy=aggregation_strategy)
|
33 |
+
for result in results:
|
34 |
+
result["entity"] = result["entity_group"]
|
35 |
+
return {"text": text, "entities": results}
|
36 |
+
|
37 |
+
ner_models = [
|
38 |
+
"livinNector/TryNER-500",
|
39 |
+
"livinNector/TryNER-1k",
|
40 |
+
"livinNector/IndicBERTNER",
|
41 |
+
"livinNector/IndicNER",
|
42 |
+
"ai4bharat/IndicNER",
|
43 |
+
"livinNector/distilbert-multilingual-base-ner"
|
44 |
+
]
|
45 |
+
ner_pipes = [pipeline("token-classification",model) for model in ner_models]
|
46 |
+
|
47 |
+
def get_ner_outputs(text,aggregation_strategy):
|
48 |
+
return [get_ner(pipe,text,aggregation_strategy) for pipe in ner_pipes]
|
49 |
+
examples = [
|
50 |
+
["ஆனந்த் மற்றும் லிவின் நெக்டர் ஆகியொர் அண்ணாமலை பல்கலைக்கழகத்தில் படித்து வருகின்றனர்.","first"],
|
51 |
+
["இந்தியன் இன்ஸ்டிட்யூட் ஆஃப் டெக்னாலஜி மெட்ராஸ் கிண்டியில் அமைந்துள்ளது.","average"],
|
52 |
+
["சச்சின் டெண்டுல்கர் மும்பை மாநகரத்தைச் சேர்ந்த ஒரு நடுத்தரக் குடும்பத்தில் நான்காவது குழந்தையாகப் பிறந்தார். பல துடுப்பாட்ட வீரர்களை உருவாக்கிய சாரதாஷ்ரம் வித்யாமந்திர் பள்ளியில் சேர்ந்தார்.","bio_first"]
|
53 |
+
|
54 |
+
]
|
55 |
|
56 |
+
iface = gr.Interface(
|
57 |
+
get_ner_outputs,
|
58 |
+
[
|
59 |
+
gr.Textbox(value=examples[0][0]),
|
60 |
+
gr.Dropdown(["bio_first","first","max","average"],value=examples[0][1])
|
61 |
+
],
|
62 |
+
[gr.Highlight(label=model) for model in ner_models],
|
63 |
+
description='Named Entity Recongnition Interface Comparing Various Transformer Based NER models for Tamil Language.',
|
64 |
+
examples=examples,
|
65 |
+
title='TaNER',
|
66 |
+
|
67 |
+
)
|
68 |
|
69 |
+
iface.launch(enable_queue=True)
|