Update app.py

app.py

@@ -1,31 +1,30 @@
 import gradio as gr
-from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+from transformers import T5Tokenizer, T5ForConditionalGeneration
 
-model_name = "ramsrigouthamg/t5_paraphraser"
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
+# Load pretrained T5 model for paraphrasing
+model_name = "Vamsi/T5_Paraphrase_Paws"
+tokenizer = T5Tokenizer.from_pretrained(model_name)
+model = T5ForConditionalGeneration.from_pretrained(model_name)
 
-def paraphrase_text(text):
-    input_text = f"paraphrase: {text} </s>"
-    encoding = tokenizer.encode_plus(input_text, padding=True, return_tensors="pt")
-    outputs = model.generate(
-        input_ids=encoding["input_ids"],
-        attention_mask=encoding["attention_mask"],
-        max_length=256,
-        num_return_sequences=1,
-        do_sample=True,
-        top_k=120,
-        top_p=0.95,
-        early_stopping=True
-    )
-    return tokenizer.decode(outputs[0], skip_special_tokens=True)
+def paraphrase_text(input_text):
+    # Prepare input text
+    input_ids = tokenizer.encode("paraphrase: " + input_text + " </s>", return_tensors="pt", max_length=512, truncation=True)
+    
+    # Generate paraphrase
+    outputs = model.generate(input_ids, max_length=256, num_return_sequences=1, temperature=1.5)
+    
+    # Decode output
+    paraphrased_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return paraphrased_text
 
-iface = gr.Interface(
+# Create Gradio interface
+demo = gr.Interface(
     fn=paraphrase_text,
-    inputs=gr.Textbox(lines=8, label="Enter Text"),
-    outputs=gr.Textbox(label="Paraphrased Text"),
-    title="AI Paraphraser Tool",
-    description="This tool uses T5 to paraphrase your input text."
+    inputs=gr.Textbox(lines=5, placeholder="Enter text to paraphrase..."),
+    outputs="text",
+    title="AI Paraphraser",
+    description="Paste any English text to get a paraphrased version using T5 model."
 )
 
-iface.launch()
+# Expose /run/predict endpoint
+demo.launch()