app.py

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

# 1️⃣ Modell & Tokenizer laden
model_name = "allenai/scibert_scivocab_uncased"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name, num_labels=3)

# 2️⃣ Dataset laden (mit spezifischer Konfiguration: "arxiv" oder "pubmed")
dataset = load_dataset("armanc/scientific_papers", "arxiv", trust_remote_code=True)  # Oder "pubmed"

# 3️⃣ Tokenisierung der Texte
def tokenize_function(examples):
    return tokenizer(examples["text"], padding="max_length", truncation=True)

tokenized_datasets = dataset.map(tokenize_function, batched=True)

# 4️⃣ Trainingsparameter setzen
training_args = TrainingArguments(
    output_dir="./results",
    evaluation_strategy="epoch",
    save_strategy="epoch",
    per_device_train_batch_size=8,
    per_device_eval_batch_size=8,
    num_train_epochs=3,
    weight_decay=0.01,
    logging_dir="./logs",
)

# 5️⃣ Training starten
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_datasets["train"],
    eval_dataset=tokenized_datasets["validation"],
)

trainer.train()

# 6️⃣ Speichern des Modells nach dem Training
model.save_pretrained("./trained_model")
tokenizer.save_pretrained("./trained_model")

print(dataset)  # Zeigt die Struktur des Datensatzes
print("✅ Training abgeschlossen! Modell gespeichert.")