Update app.py

app.py

@@ -13,15 +13,14 @@ pipeline = PIPELINE(model, "rwkv_vocab_v20230424")
 
 
 def generate_prompt(instruction, input=None, history=None):
-    # parse the chat history into a string of user and assistant messages
-    history_str = ""
-
     has_history = (history is not None)
     if has_history:
+        input = "\""
         for pair in history:
             if len(pair[0]) > 0 and len(pair[1]) > 0:
-                history_str += f"User: {pair[0]}\n\nAssistant: {pair[1]}\n\n"
-        
+                input += f"User: {pair[0]}\nAssistant: {pair[1]}\n"
+        input = input[:-1] + "\"\n" + f"User: {instruction}"
+        instruction = "Based on the Conversation History, generate a Response for the User."
     if instruction:
         instruction = (
             instruction.strip()
@@ -29,21 +28,20 @@ def generate_prompt(instruction, input=None, history=None):
             .replace("\n\n", "\n")
             .replace("\n\n", "\n")
         )
-    if input:
+    if input and len(input) > 0:
         input = (
             input.strip()
             .replace("\r\n", "\n")
             .replace("\n\n", "\n")
             .replace("\n\n", "\n")
         )
-    if not has_history and len(input) > 0:
-        return f"""{history_str}Instruction: {instruction}
+        return f"""Instruction: {instruction}
 
 Input: {input}
 
 Response:"""
     else:
-        return f"""{history_str}User: {instruction}
+        return f"""User: {instruction}
 
 Assistant:"""
 
@@ -177,7 +175,11 @@ def alternative(chatbot, history):
 
 with gr.Blocks(title=title) as demo:
     gr.HTML(f'<div style="text-align: center;">\n<h1>🌍World - {title}</h1>\n</div>')
-
+    gr.Markdown(
+        f"100% RNN RWKV-LM **trained on 100+ natural languages**. Demo limited to ctxlen {ctx_limit}. For best results, <b>keep your prompt short and clear</b>."
+        + "\n\n"
+        + f"Clone this space for faster inference if you can run the app on GPU or better CPU. To use CUDA, replace <code>strategy='cpu bf16'</code> with <code>strategy='cuda fp16'</code> in `app.py`."
+    )
     with gr.Tab("Chat mode"):
         with gr.Row():
             with gr.Column():
@@ -243,9 +245,6 @@ with gr.Blocks(title=title) as demo:
             )
 
     with gr.Tab("Instruct mode"):
-        gr.Markdown(
-            f"100% RNN RWKV-LM **trained on 100+ natural languages**. Demo limited to ctxlen {ctx_limit}. For best results, <b>keep your prompt short and clear</b>."
-        )
         with gr.Row():
             with gr.Column():
                 instruction = gr.Textbox(