Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import XLMRobertaTokenizer, XLMRobertaForSequenceClassification, Trainer, TrainingArguments
|
2 |
+
import datasets
|
3 |
+
|
4 |
+
# Load the pre-trained XLM-Roberta-Large model and tokenizer
|
5 |
+
model_name = 'xlm-roberta-large'
|
6 |
+
tokenizer = XLMRobertaTokenizer.from_pretrained(model_name)
|
7 |
+
model = XLMRobertaForSequenceClassification.from_pretrained(model_name, num_labels=2)
|
8 |
+
|
9 |
+
# Load the sentiment analysis dataset
|
10 |
+
dataset = datasets.load_dataset('imdb')
|
11 |
+
|
12 |
+
# Tokenize the dataset
|
13 |
+
def tokenize(batch):
|
14 |
+
return tokenizer(batch['text'], padding=True, truncation=True)
|
15 |
+
|
16 |
+
dataset = dataset.map(tokenize, batched=True)
|
17 |
+
|
18 |
+
# Fine-tune the model on the dataset
|
19 |
+
training_args = TrainingArguments(
|
20 |
+
output_dir='./results',
|
21 |
+
evaluation_strategy='epoch',
|
22 |
+
learning_rate=2e-5,
|
23 |
+
per_device_train_batch_size=8,
|
24 |
+
per_device_eval_batch_size=8,
|
25 |
+
num_train_epochs=3,
|
26 |
+
weight_decay=0.01,
|
27 |
+
push_to_hub=False,
|
28 |
+
logging_dir='./logs',
|
29 |
+
logging_steps=10,
|
30 |
+
load_best_model_at_end=True,
|
31 |
+
metric_for_best_model='accuracy'
|
32 |
+
)
|
33 |
+
|
34 |
+
trainer = Trainer(
|
35 |
+
model=model,
|
36 |
+
args=training_args,
|
37 |
+
train_dataset=dataset['train'],
|
38 |
+
eval_dataset=dataset['test']
|
39 |
+
)
|
40 |
+
|
41 |
+
trainer.train()
|
42 |
+
|
43 |
+
import torch
|
44 |
+
|
45 |
+
# Load the fine-tuned XLM-Roberta-Large model
|
46 |
+
model_path = './results/checkpoint-1000'
|
47 |
+
model = XLMRobertaForSequenceClassification.from_pretrained(model_path)
|
48 |
+
|
49 |
+
# Create a function that takes a text input and returns the predicted sentiment label
|
50 |
+
def predict_sentiment(text):
|
51 |
+
inputs = tokenizer(text, padding=True, truncation=True, return_tensors='pt')
|
52 |
+
outputs = model(**inputs)
|
53 |
+
logits = outputs.logits
|
54 |
+
predicted_class = torch.argmax(logits, dim=1)
|
55 |
+
return 'positive' if predicted_class == 1 else 'negative'
|
56 |
+
|
57 |
+
import gradio as gr
|
58 |
+
|
59 |
+
# Create a Gradio interface for the predict_sentiment function
|
60 |
+
iface = gr.Interface(
|
61 |
+
fn=predict_sentiment,
|
62 |
+
inputs=gr.inputs.Textbox(placeholder='Enter text here...'),
|
63 |
+
outputs=gr.outputs.Textbox(placeholder='Sentiment prediction...')
|
64 |
+
)
|
65 |
+
|
66 |
+
# Launch the interface
|
67 |
+
iface.launch()
|