Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -5,6 +5,7 @@ from transformers import TFAutoModelForSequenceClassification, AutoTokenizer
|
|
5 |
|
6 |
model = TFAutoModelForSequenceClassification.from_pretrained("bert-base-uncased")
|
7 |
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
|
|
|
8 |
|
9 |
ds = load_dataset("stanfordnlp/sst2")
|
10 |
|
@@ -18,6 +19,39 @@ def encode(examples):
|
|
18 |
sst2_dataset = sst2_dataset.map(encode, batched=True)
|
19 |
sst2_dataset = sst2_dataset.map(lambda examples: {"labels": examples["label"]}, batched=True)
|
20 |
|
21 |
-
|
22 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
|
|
|
5 |
|
6 |
model = TFAutoModelForSequenceClassification.from_pretrained("bert-base-uncased")
|
7 |
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
|
8 |
+
model = BertForSequenceClassification.from_pretrained(model_name, num_labels=2) # 2 classes : positif et négatif
|
9 |
|
10 |
ds = load_dataset("stanfordnlp/sst2")
|
11 |
|
|
|
19 |
sst2_dataset = sst2_dataset.map(encode, batched=True)
|
20 |
sst2_dataset = sst2_dataset.map(lambda examples: {"labels": examples["label"]}, batched=True)
|
21 |
|
22 |
+
training_args = TrainingArguments(
|
23 |
+
per_device_train_batch_size=8,
|
24 |
+
evaluation_strategy="epoch",
|
25 |
+
logging_dir="./logs",
|
26 |
+
output_dir="./results",
|
27 |
+
num_train_epochs=3,
|
28 |
+
)
|
29 |
+
|
30 |
+
|
31 |
+
trainer = Trainer(
|
32 |
+
model=model,
|
33 |
+
args=training_args,
|
34 |
+
train_dataset=encoded_dataset["train"],
|
35 |
+
eval_dataset=encoded_dataset["test"],
|
36 |
+
)
|
37 |
+
|
38 |
+
import os
|
39 |
+
if not os.path.exists("./fine_tuned_model"):
|
40 |
+
trainer.train()
|
41 |
+
# Sauvegarder le modèle fine-tuné et le tokenizer
|
42 |
+
model.save_pretrained("./fine_tuned_model")
|
43 |
+
tokenizer.save_pretrained("./fine_tuned_model")
|
44 |
+
else:
|
45 |
+
# Charger le modèle fine-tuné
|
46 |
+
model = BertForSequenceClassification.from_pretrained("./fine_tuned_model")
|
47 |
+
tokenizer = BertTokenizer.from_pretrained("./fine_tuned_model")
|
48 |
+
|
49 |
+
|
50 |
+
sentiment_analysis = pipeline("sentiment-analysis", model=model, tokenizer=tokenizer)
|
51 |
+
|
52 |
+
def generate_response(message):
|
53 |
+
result = sentiment_analysis(message)[0]
|
54 |
+
return f"Label: {result['label']}, Score: {result['score']}"
|
55 |
+
|
56 |
+
gr.ChatInterface(fn=generate_response).launch()
|
57 |
|