app.py

import datasets
from datasets import load_dataset
import transformers 
from transformers import AutoModelForSequenceClassification, AutoTokenizer, Trainer, TrainingArguments, pipeline

model = AutoModelForSequenceClassification.from_pretrained("bert-base-uncased", num_labels=2)  # 2 classes : positif et négatif
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")

ds = load_dataset("stanfordnlp/sst2")

sst2_dataset = load_dataset("glue", "sst2", split="train")


def encode(examples):
    return tokenizer(examples["sentence1"], examples["sentence2"], truncation=True, padding="max_length")


sst2_dataset = sst2_dataset.map(encode, batched=True)
sst2_dataset = sst2_dataset.map(lambda examples: {"labels": examples["label"]}, batched=True)

training_args = TrainingArguments(
    per_device_train_batch_size=8,
    evaluation_strategy="epoch",
    logging_dir="./logs",
    output_dir="./results",
    num_train_epochs=3,
)


trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=encoded_dataset["train"],
    eval_dataset=encoded_dataset["test"],
)

import os
if not os.path.exists("./fine_tuned_model"):
    trainer.train()
    # Sauvegarder le modèle fine-tuné et le tokenizer
    model.save_pretrained("./fine_tuned_model")
    tokenizer.save_pretrained("./fine_tuned_model")
else:
    # Charger le modèle fine-tuné
    model = BertForSequenceClassification.from_pretrained("./fine_tuned_model")
    tokenizer = BertTokenizer.from_pretrained("./fine_tuned_model")


sentiment_analysis = pipeline("sentiment-analysis", model=model, tokenizer=tokenizer)

def generate_response(message):
    result = sentiment_analysis(message)[0]
    return f"Label: {result['label']}, Score: {result['score']}"

gr.ChatInterface(fn=generate_response).launch()