Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -4,16 +4,16 @@ from transformers import MarianTokenizer
|
|
4 |
import gradio as gr
|
5 |
|
6 |
# Load the tokenizer from the local folder
|
7 |
-
|
8 |
-
tokenizer = MarianTokenizer.from_pretrained(
|
9 |
|
10 |
# Load the ONNX model
|
11 |
onnx_model_path = "./model.onnx"
|
12 |
session = ort.InferenceSession(onnx_model_path)
|
13 |
|
14 |
-
def
|
15 |
# Tokenize the input texts
|
16 |
-
inputs = tokenizer(
|
17 |
input_ids = inputs["input_ids"].astype(np.int64)
|
18 |
attention_mask = inputs["attention_mask"].astype(np.int64)
|
19 |
|
@@ -24,7 +24,7 @@ def translate_text(input_texts, max_length=512):
|
|
24 |
# Generate output tokens iteratively
|
25 |
for _ in range(max_length):
|
26 |
# Run the ONNX model
|
27 |
-
|
28 |
None,
|
29 |
{
|
30 |
"input_ids": input_ids,
|
@@ -34,7 +34,7 @@ def translate_text(input_texts, max_length=512):
|
|
34 |
)
|
35 |
|
36 |
# Get the next token logits (output of the ONNX model)
|
37 |
-
next_token_logits =
|
38 |
|
39 |
# Greedy decoding: select the token with the highest probability
|
40 |
next_tokens = np.argmax(next_token_logits, axis=-1) # Shape: (batch_size,)
|
@@ -51,17 +51,20 @@ def translate_text(input_texts, max_length=512):
|
|
51 |
return translations
|
52 |
|
53 |
# Gradio interface
|
54 |
-
def gradio_translate(
|
55 |
-
|
56 |
-
|
|
|
|
|
|
|
57 |
|
58 |
# Create the Gradio interface
|
59 |
interface = gr.Interface(
|
60 |
fn=gradio_translate,
|
61 |
-
inputs=gr.Textbox(lines=
|
62 |
-
outputs=gr.Textbox(label="Translated Text"),
|
63 |
title="ONNX English to French Translation",
|
64 |
-
description="Translate English text to French using
|
65 |
)
|
66 |
|
67 |
# Launch the Gradio app
|
|
|
4 |
import gradio as gr
|
5 |
|
6 |
# Load the tokenizer from the local folder
|
7 |
+
tokenizer_path = "./onnx_model" # Path to the local tokenizer folder
|
8 |
+
tokenizer = MarianTokenizer.from_pretrained(tokenizer_path)
|
9 |
|
10 |
# Load the ONNX model
|
11 |
onnx_model_path = "./model.onnx"
|
12 |
session = ort.InferenceSession(onnx_model_path)
|
13 |
|
14 |
+
def translate(texts, max_length=512):
|
15 |
# Tokenize the input texts
|
16 |
+
inputs = tokenizer(texts, return_tensors="np", padding=True, truncation=True, max_length=max_length)
|
17 |
input_ids = inputs["input_ids"].astype(np.int64)
|
18 |
attention_mask = inputs["attention_mask"].astype(np.int64)
|
19 |
|
|
|
24 |
# Generate output tokens iteratively
|
25 |
for _ in range(max_length):
|
26 |
# Run the ONNX model
|
27 |
+
onnx_outputs = session.run(
|
28 |
None,
|
29 |
{
|
30 |
"input_ids": input_ids,
|
|
|
34 |
)
|
35 |
|
36 |
# Get the next token logits (output of the ONNX model)
|
37 |
+
next_token_logits = onnx_outputs[0][:, -1, :] # Shape: (batch_size, vocab_size)
|
38 |
|
39 |
# Greedy decoding: select the token with the highest probability
|
40 |
next_tokens = np.argmax(next_token_logits, axis=-1) # Shape: (batch_size,)
|
|
|
51 |
return translations
|
52 |
|
53 |
# Gradio interface
|
54 |
+
def gradio_translate(input_text):
|
55 |
+
# Split the input text into lines (assuming one sentence per line)
|
56 |
+
texts = input_text.strip().split("\n")
|
57 |
+
translations = translate(texts)
|
58 |
+
# Join the translations into a single string with line breaks
|
59 |
+
return "\n".join(translations)
|
60 |
|
61 |
# Create the Gradio interface
|
62 |
interface = gr.Interface(
|
63 |
fn=gradio_translate,
|
64 |
+
inputs=gr.Textbox(lines=5, placeholder="Enter text to translate...", label="Input Text"),
|
65 |
+
outputs=gr.Textbox(lines=5, label="Translated Text"),
|
66 |
title="ONNX English to French Translation",
|
67 |
+
description="Translate English text to French using an ONNX model.",
|
68 |
)
|
69 |
|
70 |
# Launch the Gradio app
|