Create paraphraser.py

paraphraser.py

@@ -0,0 +1,33 @@
+# paraphraser.py
+import torch
+from model_loader import paraphrase_model, paraphrase_tokenizer
+
+def paraphrase_comment(comment, prompt_template="Paraphrase the following comment to make it non-toxic while preserving its meaning: \"{comment}\""):
+    """
+    Paraphrase a toxic comment using the Granite 3.2-2B-Instruct model.
+    Returns the paraphrased comment.
+    """
+    # Format the prompt with the input comment
+    prompt = prompt_template.format(comment=comment)
+
+    # Tokenize the prompt
+    inputs = paraphrase_tokenizer(prompt, return_tensors="pt", truncation=True, padding=True, max_length=512)
+
+    # Generate the paraphrased output
+    with torch.no_grad():
+        outputs = paraphrase_model.generate(
+            **inputs,
+            max_length=512,
+            num_return_sequences=1,
+            do_sample=True,
+            top_p=0.95,
+            temperature=0.7
+        )
+
+    # Decode the generated output
+    paraphrased_comment = paraphrase_tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+    # Remove the prompt part from the output (if the model includes it)
+    paraphrased_comment = paraphrased_comment.replace(prompt, "").strip()
+
+    return paraphrased_comment