inputs = tokenizer(formatted_prompt, return_tensors="pt").to(model.device)

with torch.no_grad():
    output = model.generate(
        inputs["input_ids"],
        max_new_tokens=max_new_tokens,
        temperature=0.7,
        top_p=0.9,
        do_sample=True
    )

# Decode and extract only the generated part (not the prompt)
full_output = tokenizer.decode(output[0], skip_special_tokens=True)
generated_text = full_output[len(tokenizer.decode(inputs["input_ids"][0], skip_special_tokens=True)):]

return generated_text.strip()

# Correctly extract product and description from the current sample
product = sample["product"]
description = sample["description"]

# Create the prompt
prompt = f"Generate a marketing email with the subject line below:\n Product: {product}\nDescription: {description}\n\n"
reference = sample["marketing_email"]

# Generate text using the model
generated_text = generate_text(prompt)

# Calculate ROUGE scores
scores = scorer.score(reference, generated_text)

# Store results
results.append({
    "index": i,
    "prompt": prompt,
    "reference": reference,
    "generated": generated_text,
    "rouge1_precision": scores["rouge1"].precision,
    "rouge1_recall": scores["rouge1"].recall,
    "rouge1_fmeasure": scores["rouge1"].fmeasure,
    "rouge2_precision": scores["rouge2"].precision,
    "rouge2_recall": scores["rouge2"].recall,
    "rouge2_fmeasure": scores["rouge2"].fmeasure,
    "rougeL_precision": scores["rougeL"].precision,
    "rougeL_recall": scores["rougeL"].recall,
    "rougeL_fmeasure": scores["rougeL"].fmeasure

})