Spaces:
Sleeping
Sleeping
Commit
·
fc36581
1
Parent(s):
89b5af7
Update app.py
Browse files
app.py
CHANGED
@@ -1,65 +1,37 @@
|
|
1 |
import numpy as np
|
2 |
import onnxruntime as ort
|
3 |
-
|
|
|
4 |
import gradio as gr
|
5 |
|
6 |
-
# Load the
|
7 |
-
model_path = "model.onnx"
|
8 |
-
|
9 |
-
|
|
|
|
|
|
|
10 |
|
11 |
def translate_text(input_text):
|
12 |
# Tokenize input text
|
13 |
-
tokenized_input =
|
14 |
-
input_text, return_tensors="
|
15 |
)
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
for _ in range(512): # Max length of output
|
28 |
-
# Run inference with the ONNX model
|
29 |
-
outputs = translation_session.run(
|
30 |
-
None,
|
31 |
-
{
|
32 |
-
"input_ids": input_ids,
|
33 |
-
"attention_mask": attention_mask,
|
34 |
-
"decoder_input_ids": decoder_input_ids,
|
35 |
-
}
|
36 |
-
)
|
37 |
-
|
38 |
-
# Get the next token ID
|
39 |
-
next_token_id = np.argmax(outputs[0][0, -1, :], axis=-1)
|
40 |
-
translated_tokens.append(next_token_id)
|
41 |
-
|
42 |
-
# Stop if the end-of-sequence token is generated
|
43 |
-
if next_token_id == translation_tokenizer.eos_token_id:
|
44 |
-
break
|
45 |
-
|
46 |
-
# Update decoder_input_ids for the next iteration
|
47 |
-
decoder_input_ids = np.concatenate(
|
48 |
-
[decoder_input_ids, np.array([[next_token_id]], dtype=np.int64)], axis=1
|
49 |
)
|
50 |
|
51 |
# Decode the output tokens
|
52 |
-
translated_text =
|
53 |
return translated_text
|
54 |
|
55 |
-
# Create a Gradio interface
|
56 |
-
interface = gr.Interface(
|
57 |
-
fn=translate_text,
|
58 |
-
inputs="text",
|
59 |
-
outputs="text",
|
60 |
-
title="Frenchizer Translation Model",
|
61 |
-
description="Translate text from English to French using an ONNX model."
|
62 |
-
)
|
63 |
-
|
64 |
-
# Launch the Gradio app
|
65 |
interface.launch()
|
|
|
1 |
import numpy as np
|
2 |
import onnxruntime as ort
|
3 |
+
import torch
|
4 |
+
from transformers import MarianMTModel, MarianTokenizer
|
5 |
import gradio as gr
|
6 |
|
7 |
+
# Load the MarianMT model and tokenizer from the local folder
|
8 |
+
model_path = "./model.onnx" # Path to the folder containing the model files
|
9 |
+
tokenizer = MarianTokenizer.from_pretrained(model_name)
|
10 |
+
decoder_model = MarianMTModel.from_pretrained(model_name).get_decoder()
|
11 |
+
|
12 |
+
# Load the ONNX encoder
|
13 |
+
encoder_session = ort.InferenceSession("./onnx_model/encoder.onnx")
|
14 |
|
15 |
def translate_text(input_text):
|
16 |
# Tokenize input text
|
17 |
+
tokenized_input = tokenizer(
|
18 |
+
input_text, return_tensors="pt", padding=True, truncation=True, max_length=512
|
19 |
)
|
20 |
+
input_ids = tokenized_input["input_ids"]
|
21 |
+
attention_mask = tokenized_input["attention_mask"]
|
22 |
+
|
23 |
+
# Generate translation using the model
|
24 |
+
with torch.no_grad():
|
25 |
+
outputs = model.generate(
|
26 |
+
input_ids=input_ids,
|
27 |
+
attention_mask=attention_mask,
|
28 |
+
max_length=512, # Maximum length of the output
|
29 |
+
num_beams=5, # Use beam search for better translations
|
30 |
+
early_stopping=True, # Stop generation when the model predicts the end-of-sequence token
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
)
|
32 |
|
33 |
# Decode the output tokens
|
34 |
+
translated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
35 |
return translated_text
|
36 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
37 |
interface.launch()
|