Update app.py

app.py

@@ -39,6 +39,28 @@ def generate_text(prompt, max_new_tokens=128):
     generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
     return generated_text
 
+def generate_text_streaming(prompt, max_new_tokens=128):
+    input_ids = tokenizer.encode(prompt, return_tensors="pt").to(model.device)
+
+    with torch.no_grad():
+        for i in range(max_new_tokens):
+            output = model.generate(
+                input_ids=input_ids,
+                max_new_tokens=1,  # Generate only 1 new token at a time
+                do_sample=False,  # Or True for sampling
+                eos_token_id=tokenizer.eos_token_id,
+                return_dict=True, #Return a dictionary
+                output_scores=True #Return the scores
+            )
+
+            generated_token = tokenizer.decode(output.logits[0][-1].argmax(), skip_special_tokens=True) #Decode the last token only
+            yield generated_token #Yield the last token
+
+            input_ids = torch.cat([input_ids, output.sequences[:, -1:]], dim=-1) #Append the new token to the input
+
+            if output.sequences[0][-1] == tokenizer.eos_token_id: #Check if the end of sequence token was generated
+                break #Break the loop
+
 def respond(message, history, system_message, max_tokens):
     prompt = f"{system_message}\n"
     for user_msg, bot_msg in history:
@@ -46,8 +68,9 @@ def respond(message, history, system_message, max_tokens):
     prompt += f"User: {message}\nAssistant:"
 
     try:
-        bot_response = generate_text(prompt, max_tokens)  # Use the new function
-        yield bot_response
+        for token in generate_text_streaming(prompt, max_tokens): #Iterate over the generator
+            yield token #Yield each token individually
+
     except Exception as e:
         print(f"Error during generation: {e}")
         yield "An error occurred."
@@ -59,7 +82,7 @@ demo = gr.ChatInterface(
             value="You are a friendly and helpful mental health chatbot.",
             label="System message",
         ),
-        gr.Slider(minimum=1, maximum=128, value=128, step=10, label="Max new tokens"),
+        gr.Slider(minimum=1, maximum=128, value=32, step=10, label="Max new tokens"),
     ],
 )