Update generate_text function with attention_mask and pad_token_id

script_analyzer.py

@@ -11,10 +11,19 @@ sentiment_classifier = pipeline("sentiment-analysis", model=model_sentiment, tok
 tokenizer_gpt = AutoTokenizer.from_pretrained("EleutherAI/gpt-neo-2.7B")
 model_gpt = AutoModelForCausalLM.from_pretrained("EleutherAI/gpt-neo-2.7B")
 
-# Helper function to generate text using GPT-Neo or GPT-J
+# Updated generate_text function with attention_mask and pad_token_id
 def generate_text(prompt):
-    inputs = tokenizer_gpt(prompt, return_tensors="pt", truncation=True, max_length=50)
-    outputs = model_gpt.generate(inputs["input_ids"], max_length=50, num_return_sequences=1, no_repeat_ngram_size=2)
+    # Prepare the input tensors with attention_mask and padding
+    inputs = tokenizer_gpt(prompt, return_tensors="pt", padding=True, truncation=True, max_length=50)
+    # Generate text using max_new_tokens instead of max_length
+    outputs = model_gpt.generate(
+        inputs["input_ids"],
+        attention_mask=inputs["attention_mask"],
+        max_new_tokens=50,  # Controls the new tokens generated beyond input length
+        num_return_sequences=1,
+        no_repeat_ngram_size=2,
+        pad_token_id=tokenizer_gpt.eos_token_id  # Sets padding to eos_token_id to prevent issues
+    )
     generated_text = tokenizer_gpt.decode(outputs[0], skip_special_tokens=True)
     return generated_text.strip()
 
@@ -114,5 +123,5 @@ with gr.Blocks() as interface:
     
     display_dashboard_button.click(display_dashboard, inputs=script_input, outputs=[output_dashboard, output_graph])
 
-# Launch the Gradio app with sharing enabled
-interface.launch(share=True)
+# Launch the Gradio app (no need for share=True in Hugging Face Spaces)
+interface.launch()