Update app.py

app.py

@@ -2,19 +2,43 @@ import gradio as gr
 from transformers import AutoModelForQuestionAnswering, AutoTokenizer, pipeline
 import json
 
+# Define hyperparameters
+learning_rate = 3e-5
+batch_size = 16
+epochs = 3
+max_seq_length = 512
+warmup_steps = 100
+weight_decay = 0.01
+dropout_prob = 0.1
+gradient_clip_value = 1.0
+
 context_val = ''
 
 q_n_a_model_name = "deepset/roberta-base-squad2"
 q_n_a_model = AutoModelForQuestionAnswering.from_pretrained(q_n_a_model_name)
-tokenizer = AutoTokenizer.from_pretrained(q_n_a_model_name)  # Corrected this line
+tokenizer = AutoTokenizer.from_pretrained(q_n_a_model_name)  
 
 context = gr.Textbox(label="Add the Context (Paragraph or texts) for which you want to get insights", lines=10, outputs="text")
 
 def q_n_a_fn(context, text):
     QA_input = {'question': text, 'context': context}
     nlp = pipeline('question-answering', model=q_n_a_model, tokenizer=tokenizer)
-    res = nlp(QA_input)
-    answer = res['answer']
+
+    # Set the device (CPU or GPU)
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    q_n_a_model.to(device)
+    
+    # Convert inputs to tensors
+    inputs = tokenizer(QA_input["context"], QA_input["question"], return_tensors="pt", max_length=max_seq_length, truncation=True, padding="max_length").to(device)
+
+    # Get predictions
+    with torch.no_grad():
+        outputs = q_n_a_model(**inputs)
+    
+    # Decode and return the answer
+    start_idx, end_idx = torch.argmax(outputs.start_logits), torch.argmax(outputs.end_logits)
+    answer = tokenizer.decode(inputs["input_ids"][0][start_idx:end_idx+1])
+    
     return answer
 
 def classification_fn(text):
@@ -51,4 +75,4 @@ with gr.Blocks(theme='gradio/soft') as demo:
             gr.Interface(fn=classification_fn, inputs=[context], outputs="text")
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()