Add system message

app.py

@@ -1,68 +1,96 @@
 import openai
 import tiktoken
+
+import datetime
 import json
 import os
 
 openai.api_key = os.getenv('API_KEY')
+openai.request_times = 0
 
-def ask(question, history):
-    history = history + [question]
+def ask(question, history, behavior):
+    openai.request_times += 1
+    print(f"request times {openai.request_times}: {datetime.datetime.now()}")
     try:
         response = openai.ChatCompletion.create(
             model="gpt-3.5-turbo",
-            messages=forget_long_term([
-                {"role":"user" if i%2==0 else "assistant", "content":content}
-                for i,content in enumerate(history)
-            ])
+            messages=forget_long_term(
+                [
+                    {"role":"system", "content":content}
+                    for content in behavior
+                ] + [
+                    {"role":"user" if i%2==0 else "assistant", "content":content}
+                    for i,content in enumerate(history + [question])
+                ]
+            )
         )["choices"][0]["message"]["content"]
         while response.startswith("\n"):
             response = response[1:]
     except Exception as e:
         print(e)
         response = 'Timeout! Please wait a few minutes and retry'
-    history = history + [response]
-    with open("dialogue.txt", "a", encoding='utf-8') as f:
-        f.write(json.dumps(history, ensure_ascii=False)+"\n")
+    history = history + [question, response]
     return history
 
+def num_tokens_from_messages(messages, model="gpt-3.5-turbo"):
+    """Returns the number of tokens used by a list of messages."""
+    try:
+        encoding = tiktoken.encoding_for_model(model)
+    except KeyError:
+        encoding = tiktoken.get_encoding("cl100k_base")
+    if model == "gpt-3.5-turbo":  # note: future models may deviate from this
+        num_tokens = 0
+        for message in messages:
+            num_tokens += 4  # every message follows <im_start>{role/name}\n{content}<im_end>\n
+            for key, value in message.items():
+                num_tokens += len(encoding.encode(value))
+                if key == "name":  # if there's a name, the role is omitted
+                    num_tokens += -1  # role is always required and always 1 token
+        num_tokens += 2  # every reply is primed with <im_start>assistant
+        return num_tokens
+    else:
+        raise NotImplementedError(f"""num_tokens_from_messages() is not presently implemented for model {model}.
+See https://github.com/openai/openai-python/blob/main/chatml.md for information on how messages are converted to tokens.""")
 
 def forget_long_term(messages, max_num_tokens=4000):
-    def num_tokens_from_messages(messages, model="gpt-3.5-turbo"):
-        """Returns the number of tokens used by a list of messages."""
-        try:
-            encoding = tiktoken.encoding_for_model(model)
-        except KeyError:
-            encoding = tiktoken.get_encoding("cl100k_base")
-        if model == "gpt-3.5-turbo":  # note: future models may deviate from this
-            num_tokens = 0
-            for message in messages:
-                num_tokens += 4  # every message follows <im_start>{role/name}\n{content}<im_end>\n
-                for key, value in message.items():
-                    num_tokens += len(encoding.encode(value))
-                    if key == "name":  # if there's a name, the role is omitted
-                        num_tokens += -1  # role is always required and always 1 token
-            num_tokens += 2  # every reply is primed with <im_start>assistant
-            return num_tokens
-        else:
-            raise NotImplementedError(f"""num_tokens_from_messages() is not presently implemented for model {model}.
-    See https://github.com/openai/openai-python/blob/main/chatml.md for information on how messages are converted to tokens.""")
     while num_tokens_from_messages(messages)>max_num_tokens:
-        messages = messages[1:]
+        if messages[0]["role"]=="system" and not len(messages[0]["content"]>=max_num_tokens):
+            messages = messages[:1] + messages[2:]
+        else:
+            messages = messages[1:]
     return messages
 
 
 import gradio as gr
 
 
-def predict(question, history=[]):
-    history = ask(question, history)
-    response = [(history[i].replace("\n","<br>"),history[i+1].replace("\n","<br>")) for i in range(0,len(history)-1,2)]
+def to_md(content):
+    is_inside_code_block = False
+    count_backtick = 0
+    output_spans = []
+    for i in range(len(content)):
+        if content[i]=="\n" and not is_inside_code_block:
+            output_spans.append("<br>")
+        elif content[i]=="`":
+            count_backtick += 1
+            if count_backtick == 3:
+                count_backtick = 0
+                is_inside_code_block = not is_inside_code_block
+            output_spans.append(content[i])
+        else:
+            output_spans.append(content[i])
+    return "".join(output_spans)
+
+
+def predict(question, history=[], behavior=[]):
+    history = ask(question, history, behavior)
+    response = [(to_md(history[i]),to_md(history[i+1])) for i in range(0,len(history)-1,2)]
     return "", history, response
 
 
 with gr.Blocks() as demo:
     
-    examples = [
+    examples_txt = [
         ['200字介绍一下凯旋门：'],
         ['网上购物有什么小窍门？'],
         ['补全下述对三亚的介绍：\n三亚位于海南岛的最南端，是'],
@@ -72,6 +100,12 @@ with gr.Blocks() as demo:
         ['polish the following statement for a paper: In this section, we perform case study to give a more intuitive demonstration of our proposed strategies and corresponding explanation.'],
     ]
     
+    examples_bhv = [
+        "你现在是一位贴心的心理咨询师，会为我提供耐心的解答。",
+        "你现在是一名无神论者，不信奉任何宗教。",
+        f"You are a helpful assistant. Today is {datetime.date.today()}.",
+    ]
+    
     gr.Markdown(
         """
         朋友你好，
@@ -80,25 +114,31 @@ with gr.Blocks() as demo:
         
         p.s. 响应时间和问题复杂程度相关，<del>一般能在10~20秒内出结果</del>用了新的api已经提速到大约5秒内了
         """)
-
-    chatbot = gr.Chatbot()
-    state = gr.State([])
     
-    with gr.Row():
-        txt = gr.Textbox(show_label=False, placeholder="Enter text and press enter").style(container=False)
-
-    txt.submit(predict, [txt, state], [txt, state, chatbot])
+    behavior = gr.State([])
     
-    with gr.Row():
-        gen = gr.Button("Submit")
-        clr = gr.Button("Clear")
+    with gr.Column(variant="panel"):
+        with gr.Row().style(equal_height=True):
+            with gr.Column(scale=0.85):
+                bhv = gr.Textbox(show_label=False, placeholder="输入你想让ChatGPT扮演的人设").style(container=False)
+            with gr.Column(scale=0.15, min_width=0):
+                button_set = gr.Button("Set")
+        gr.Examples(examples=examples_bhv, inputs=bhv)
+    bhv.submit(fn=lambda x:(x,[x]), inputs=[bhv], outputs=[bhv, behavior])
+    button_set.click(fn=lambda x:(x,[x]), inputs=[bhv], outputs=[bhv, behavior])
     
-    gen.click(fn=predict, inputs=[txt, state], outputs=[txt, state, chatbot])
-    
-    def clear(value):
-        return [], []
-    clr.click(clear, inputs=clr, outputs=[chatbot, state])
+
+    state = gr.State([])
     
-    gr_examples = gr.Examples(examples=examples, inputs=txt)
+    with gr.Column(variant="panel"):
+        chatbot = gr.Chatbot()
+        txt = gr.Textbox(show_label=False, placeholder="输入你想让ChatGPT回答的问题").style(container=False)
+        with gr.Row():
+            button_gen = gr.Button("Submit")
+            button_clr = gr.Button("Clear")
+        gr.Examples(examples=examples_txt, inputs=txt)
+    txt.submit(predict, [txt, state, behavior], [txt, state, chatbot])
+    button_gen.click(fn=predict, inputs=[txt, state, behavior], outputs=[txt, state, chatbot])
+    button_clr.click(fn=lambda :([],[]), inputs=None, outputs=[chatbot, state])
 
 demo.launch()