Update app.py

app.py

@@ -731,6 +731,17 @@ def llama_chat_sys_input_seq_constructor(text, sys_prompt=SYSTEM_PROMPT_1, bos_t
     return f"{bos_token}{B_INST} {B_SYS} {sys_prompt} {E_SYS} {text} {E_INST}"
 
 
+def few_shot_prompt(
+    message: str,
+    history: List[Tuple[str, str]], 
+    sys_prompt=SYSTEM_PROMPT_1, 
+    bos_token=BOS_TOKEN, 
+    eos_token=EOS_TOKEN,
+    include_end_instruct=True,
+):
+    return f"{bos_token} {message}"
+
+
 def llama_chat_multiturn_sys_input_seq_constructor(
     message: str,
     history: List[Tuple[str, str]], 
@@ -1561,9 +1572,10 @@ def batch_inference(
         prompt_format_fn = llama_chat_multiturn_sys_input_seq_constructor
     elif prompt_mode == 'few-shot':
         from functools import partial
-        prompt_format_fn = partial(
-            llama_chat_multiturn_sys_input_seq_constructor, include_end_instruct=False
-        )
+        # prompt_format_fn = partial(
+        #     llama_chat_multiturn_sys_input_seq_constructor, include_end_instruct=False
+        # )
+        prompt_format_fn = few_shot_prompt
     else:
         raise gr.Error(f'Wrong mode {prompt_mode}')
 
@@ -1739,7 +1751,7 @@ def launch():
                 ["upload_chat.json", "chat", 0.2, 1024, 0.5, 0, "[STOP],[END],<s>,</s>"],
                 ["upload_few_shot.json", "few-shot", 0.2, 128, 0.5, 0, "[STOP],[END],<s>,</s>,\\n"]
             ],
-            # cache_examples=True,
+            cache_examples=False,
         )
         
         demo_chat = gr.ChatInterface(