Upload 3 files

app.py

@@ -42,14 +42,6 @@ else:
 gr.Chatbot.postprocess = postprocess
 
 with gr.Blocks(css=customCSS) as demo:
-    gr.HTML(title)
-    gr.HTML('''<center><a href="https://huggingface.co/spaces/JohnSmith9982/ChuanhuChatGPT?duplicate=true"><img src="https://bit.ly/3gLdBN6" alt="复制 Space"></a>强烈建议点击上面的按钮复制一份这个Space，在你自己的Space里运行，响应更迅速、也更安全👆</center>''')
-    with gr.Row():
-        with gr.Column(scale=4):
-            keyTxt = gr.Textbox(show_label=False, placeholder=f"在这里输入你的OpenAI API-key...",value=my_api_key, type="password", visible=not HIDE_MY_KEY).style(container=True)
-        with gr.Column(scale=1):
-            use_streaming_checkbox = gr.Checkbox(label="实时传输回答", value=True, visible=enable_streaming_option)
-    chatbot = gr.Chatbot()  # .style(color_map=("#1D51EE", "#585A5B"))
     history = gr.State([])
     token_count = gr.State([])
     promptTemplates = gr.State(load_template(get_template_names(plain=True)[0], mode=2))
@@ -57,6 +49,15 @@ with gr.Blocks(css=customCSS) as demo:
     FALSECONSTANT = gr.State(False)
     topic = gr.State("未命名对话历史记录")
 
+    gr.HTML(title)
+    with gr.Row():
+        with gr.Column():
+            keyTxt = gr.Textbox(show_label=True, placeholder=f"在这里输入你的OpenAI API-key...",value=my_api_key, type="password", visible=not HIDE_MY_KEY, label="API-Key")
+        with gr.Column():
+            with gr.Row():
+                model_select_dropdown = gr.Dropdown(label="选择模型", choices=MODELS, multiselect=False, value=MODELS[0])
+                use_streaming_checkbox = gr.Checkbox(label="实时传输回答", value=True, visible=enable_streaming_option)
+    chatbot = gr.Chatbot()  # .style(color_map=("#1D51EE", "#585A5B"))
     with gr.Row():
         with gr.Column(scale=12):
             user_input = gr.Textbox(show_label=False, placeholder="在这里输入").style(
@@ -69,8 +70,9 @@ with gr.Blocks(css=customCSS) as demo:
         delLastBtn = gr.Button("🗑️ 删除最近一条对话")
         reduceTokenBtn = gr.Button("♻️ 总结对话")
     status_display = gr.Markdown("status: ready")
-    systemPromptTxt = gr.Textbox(show_label=True, placeholder=f"在这里输入System Prompt...",
-                                 label="System prompt", value=initial_prompt).style(container=True)
+
+    systemPromptTxt = gr.Textbox(show_label=True, placeholder=f"在这里输入System Prompt...", label="System prompt", value=initial_prompt).style(container=True)
+
     with gr.Accordion(label="加载Prompt模板", open=False):
         with gr.Column():
             with gr.Row():
@@ -101,28 +103,27 @@ with gr.Blocks(css=customCSS) as demo:
     #inputs, top_p, temperature, top_k, repetition_penalty
     with gr.Accordion("参数", open=False):
         top_p = gr.Slider(minimum=-0, maximum=1.0, value=1.0, step=0.05,
-                          interactive=True, label="Top-p (nucleus sampling)",)
+                        interactive=True, label="Top-p (nucleus sampling)",)
         temperature = gr.Slider(minimum=-0, maximum=5.0, value=1.0,
                                 step=0.1, interactive=True, label="Temperature",)
-        #top_k = gr.Slider( minimum=1, maximum=50, value=4, step=1, interactive=True, label="Top-k",)
-        #repetition_penalty = gr.Slider( minimum=0.1, maximum=3.0, value=1.03, step=0.01, interactive=True, label="Repetition Penalty", )
+
     gr.Markdown(description)
 
 
-    user_input.submit(predict, [keyTxt, systemPromptTxt, history, user_input, chatbot, token_count, top_p, temperature, use_streaming_checkbox], [chatbot, history, status_display, token_count], show_progress=True)
+    user_input.submit(predict, [keyTxt, systemPromptTxt, history, user_input, chatbot, token_count, top_p, temperature, use_streaming_checkbox, model_select_dropdown], [chatbot, history, status_display, token_count], show_progress=True)
     user_input.submit(reset_textbox, [], [user_input])
 
-    submitBtn.click(predict, [keyTxt, systemPromptTxt, history, user_input, chatbot, token_count, top_p, temperature, use_streaming_checkbox], [chatbot, history, status_display, token_count], show_progress=True)
+    submitBtn.click(predict, [keyTxt, systemPromptTxt, history, user_input, chatbot, token_count, top_p, temperature, use_streaming_checkbox, model_select_dropdown], [chatbot, history, status_display, token_count], show_progress=True)
     submitBtn.click(reset_textbox, [], [user_input])
 
     emptyBtn.click(reset_state, outputs=[chatbot, history, token_count, status_display], show_progress=True)
 
-    retryBtn.click(retry, [keyTxt, systemPromptTxt, history, chatbot, token_count, top_p, temperature, use_streaming_checkbox], [chatbot, history, status_display, token_count], show_progress=True)
+    retryBtn.click(retry, [keyTxt, systemPromptTxt, history, chatbot, token_count, top_p, temperature, use_streaming_checkbox, model_select_dropdown], [chatbot, history, status_display, token_count], show_progress=True)
 
-    delLastBtn.click(delete_last_conversation, [chatbot, history, token_count, use_streaming_checkbox], [
+    delLastBtn.click(delete_last_conversation, [chatbot, history, token_count], [
                      chatbot, history, token_count, status_display], show_progress=True)
 
-    reduceTokenBtn.click(reduce_token_size, [keyTxt, systemPromptTxt, history, chatbot, token_count, top_p, temperature, use_streaming_checkbox], [chatbot, history, status_display, token_count], show_progress=True)
+    reduceTokenBtn.click(reduce_token_size, [keyTxt, systemPromptTxt, history, chatbot, token_count, top_p, temperature, use_streaming_checkbox, model_select_dropdown], [chatbot, history, status_display, token_count], show_progress=True)
 
     saveHistoryBtn.click(save_chat_history, [
                   saveFileName, systemPromptTxt, history, chatbot], None, show_progress=True)

presets.py

@@ -31,9 +31,18 @@ pre code {
 }
 """
 
-standard_error_msg = "☹️发生了错误：" # 错误信息的标准前缀
-error_retrieve_prompt = "连接超时，无法获取对话。请检查网络连接，或者API-Key是否有效。" # 获取对话时发生错误
 summarize_prompt = "请总结以上对话，不超过100字。" # 总结对话时的 prompt
+MODELS = ["gpt-3.5-turbo", "gpt-3.5-turbo-0301"] # 可选的模型
+
+# 错误信息
+standard_error_msg = "☹️发生了错误：" # 错误信息的标准前缀
+error_retrieve_prompt = "请检查网络连接，或者API-Key是否有效。" # 获取对话时发生错误
+connection_timeout_prompt = "连接超时，无法获取对话。" # 连接超时
+read_timeout_prompt = "读取超时，无法获取对话。" # 读取超时
+proxy_error_prompt = "代理错误，无法获取对话。" # 代理错误
+ssl_error_prompt = "SSL错误，无法获取对话。" # SSL 错误
+no_apikey_msg = "API key长度不是51位，请检查是否输入正确。" # API key 长度不足 51 位
+
 max_token_streaming = 3500 # 流式对话时的最大 token 数
 timeout_streaming = 15 # 流式对话时的超时时间
 max_token_all = 3500 # 非流式对话时的最大 token 数

utils.py

@@ -99,7 +99,7 @@ def construct_assistant(text):
 def construct_token_message(token, stream=False):
     return f"Token 计数: {token}"
 
-def get_response(openai_api_key, system_prompt, history, temperature, top_p, stream):
+def get_response(openai_api_key, system_prompt, history, temperature, top_p, stream, selected_model):
     headers = {
         "Content-Type": "application/json",
         "Authorization": f"Bearer {openai_api_key}"
@@ -108,7 +108,7 @@ def get_response(openai_api_key, system_prompt, history, temperature, top_p, str
     history = [construct_system(system_prompt), *history]
 
     payload = {
-        "model": "gpt-3.5-turbo",
+        "model": selected_model,
         "messages": history,  # [{"role": "user", "content": f"{inputs}"}],
         "temperature": temperature,  # 1.0,
         "top_p": top_p,  # 1.0,
@@ -124,40 +124,40 @@ def get_response(openai_api_key, system_prompt, history, temperature, top_p, str
     response = requests.post(API_URL, headers=headers, json=payload, stream=True, timeout=timeout)
     return response
 
-def stream_predict(openai_api_key, system_prompt, history, inputs, chatbot, previous_token_count, top_p, temperature):
+def stream_predict(openai_api_key, system_prompt, history, inputs, chatbot, all_token_counts, top_p, temperature, selected_model):
     def get_return_value():
-        return chatbot, history, status_text, [*previous_token_count, token_counter]
+        return chatbot, history, status_text, all_token_counts
 
     print("实时回答模式")
-    token_counter = 0
     partial_words = ""
     counter = 0
     status_text = "开始实时传输回答……"
     history.append(construct_user(inputs))
+    history.append(construct_assistant(""))
+    chatbot.append((parse_text(inputs), ""))
     user_token_count = 0
-    if len(previous_token_count) == 0:
+    if len(all_token_counts) == 0:
         system_prompt_token_count = count_token(system_prompt)
         user_token_count = count_token(inputs) + system_prompt_token_count
     else:
         user_token_count = count_token(inputs)
+    all_token_counts.append(user_token_count)
     print(f"输入token计数: {user_token_count}")
+    yield get_return_value()
     try:
-        response = get_response(openai_api_key, system_prompt, history, temperature, top_p, True)
+        response = get_response(openai_api_key, system_prompt, history, temperature, top_p, True, selected_model)
     except requests.exceptions.ConnectTimeout:
-        history.pop()
-        status_text = standard_error_msg + "连接超时，无法获取对话。" + error_retrieve_prompt
+        status_text = standard_error_msg + connection_timeout_prompt + error_retrieve_prompt
         yield get_return_value()
         return
     except requests.exceptions.ReadTimeout:
-        history.pop()
-        status_text = standard_error_msg + "读取超时，无法获取对话。" + error_retrieve_prompt
+        status_text = standard_error_msg + read_timeout_prompt + error_retrieve_prompt
         yield get_return_value()
         return
 
-    chatbot.append((parse_text(inputs), ""))
     yield get_return_value()
 
-    for chunk in response.iter_lines():
+    for chunk in tqdm(response.iter_lines()):
         if counter == 0:
             counter += 1
             continue
@@ -169,77 +169,93 @@ def stream_predict(openai_api_key, system_prompt, history, inputs, chatbot, prev
             try:
                 chunk = json.loads(chunk[6:])
             except json.JSONDecodeError:
+                print(chunk)
                 status_text = f"JSON解析错误。请重置对话。收到的内容: {chunk}"
                 yield get_return_value()
-                break
+                continue
             # decode each line as response data is in bytes
             if chunklength > 6 and "delta" in chunk['choices'][0]:
                 finish_reason = chunk['choices'][0]['finish_reason']
-                status_text = construct_token_message(sum(previous_token_count)+token_counter+user_token_count, stream=True)
+                status_text = construct_token_message(sum(all_token_counts), stream=True)
                 if finish_reason == "stop":
-                    print("生成完毕")
                     yield get_return_value()
                     break
                 try:
                     partial_words = partial_words + chunk['choices'][0]["delta"]["content"]
                 except KeyError:
-                    status_text = standard_error_msg + "API回复中找不到内容。很可能是Token计数达到上限了。请重置对话。当前Token计数: " + str(sum(previous_token_count)+token_counter+user_token_count)
+                    status_text = standard_error_msg + "API回复中找不到内容。很可能是Token计数达到上限了。请重置对话。当前Token计数: " + str(sum(all_token_counts))
                     yield get_return_value()
                     break
-                if token_counter == 0:
-                    history.append(construct_assistant(" " + partial_words))
-                else:
-                    history[-1] = construct_assistant(partial_words)
+                history[-1] = construct_assistant(partial_words)
                 chatbot[-1] = (parse_text(inputs), parse_text(partial_words))
-                token_counter += 1
+                all_token_counts[-1] += 1
                 yield get_return_value()
 
 
-def predict_all(openai_api_key, system_prompt, history, inputs, chatbot, previous_token_count, top_p, temperature):
+def predict_all(openai_api_key, system_prompt, history, inputs, chatbot, all_token_counts, top_p, temperature, selected_model):
     print("一次性回答模式")
     history.append(construct_user(inputs))
+    history.append(construct_assistant(""))
+    chatbot.append((parse_text(inputs), ""))
+    all_token_counts.append(count_token(inputs))
     try:
-        response = get_response(openai_api_key, system_prompt, history, temperature, top_p, False)
+        response = get_response(openai_api_key, system_prompt, history, temperature, top_p, False, selected_model)
     except requests.exceptions.ConnectTimeout:
-        status_text = standard_error_msg + error_retrieve_prompt
-        return chatbot, history, status_text, previous_token_count
+        status_text = standard_error_msg + connection_timeout_prompt + error_retrieve_prompt
+        return chatbot, history, status_text, all_token_counts
+    except requests.exceptions.ProxyError:
+        status_text = standard_error_msg + proxy_error_prompt + error_retrieve_prompt
+        return chatbot, history, status_text, all_token_counts
+    except requests.exceptions.SSLError:
+        status_text = standard_error_msg + ssl_error_prompt + error_retrieve_prompt
+        return chatbot, history, status_text, all_token_counts
     response = json.loads(response.text)
     content = response["choices"][0]["message"]["content"]
-    history.append(construct_assistant(content))
+    history[-1] = construct_assistant(content)
     chatbot.append((parse_text(inputs), parse_text(content)))
     total_token_count = response["usage"]["total_tokens"]
-    previous_token_count.append(total_token_count - sum(previous_token_count))
+    all_token_counts[-1] = total_token_count - sum(all_token_counts)
     status_text = construct_token_message(total_token_count)
-    print("生成一次性回答完毕")
-    return chatbot, history, status_text, previous_token_count
+    return chatbot, history, status_text, all_token_counts
 
 
-def predict(openai_api_key, system_prompt, history, inputs, chatbot, token_count, top_p, temperature, stream=False, should_check_token_count = True):  # repetition_penalty, top_k
+def predict(openai_api_key, system_prompt, history, inputs, chatbot, all_token_counts, top_p, temperature, stream=False, selected_model = MODELS[0], should_check_token_count = True):  # repetition_penalty, top_k
     print("输入为：" +colorama.Fore.BLUE + f"{inputs}" + colorama.Style.RESET_ALL)
+    if len(openai_api_key) != 51:
+        status_text = standard_error_msg + no_apikey_msg
+        print(status_text)
+        history.append(construct_user(inputs))
+        history.append("")
+        chatbot.append((parse_text(inputs), ""))
+        all_token_counts.append(0)
+        yield chatbot, history, status_text, all_token_counts
+        return
+    yield chatbot, history, "开始生成回答……", all_token_counts
     if stream:
         print("使用流式传输")
-        iter = stream_predict(openai_api_key, system_prompt, history, inputs, chatbot, token_count, top_p, temperature)
-        for chatbot, history, status_text, token_count in iter:
-            yield chatbot, history, status_text, token_count
+        iter = stream_predict(openai_api_key, system_prompt, history, inputs, chatbot, all_token_counts, top_p, temperature, selected_model)
+        for chatbot, history, status_text, all_token_counts in iter:
+            yield chatbot, history, status_text, all_token_counts
     else:
         print("不使用流式传输")
-        chatbot, history, status_text, token_count = predict_all(openai_api_key, system_prompt, history, inputs, chatbot, token_count, top_p, temperature)
-        yield chatbot, history, status_text, token_count
-    print(f"传输完毕。当前token计数为{token_count}")
-    print("回答为：" +colorama.Fore.BLUE + f"{history[-1]['content']}" + colorama.Style.RESET_ALL)
+        chatbot, history, status_text, all_token_counts = predict_all(openai_api_key, system_prompt, history, inputs, chatbot, all_token_counts, top_p, temperature, selected_model)
+        yield chatbot, history, status_text, all_token_counts
+    print(f"传输完毕。当前token计数为{all_token_counts}")
+    if len(history) > 1 and history[-1]['content'] != inputs:
+        print("回答为：" +colorama.Fore.BLUE + f"{history[-1]['content']}" + colorama.Style.RESET_ALL)
     if stream:
         max_token = max_token_streaming
     else:
         max_token = max_token_all
-    if sum(token_count) > max_token and should_check_token_count:
-        print(f"精简token中{token_count}/{max_token}")
-        iter = reduce_token_size(openai_api_key, system_prompt, history, chatbot, token_count, top_p, temperature, stream=False, hidden=True)
-        for chatbot, history, status_text, token_count in iter:
+    if sum(all_token_counts) > max_token and should_check_token_count:
+        print(f"精简token中{all_token_counts}/{max_token}")
+        iter = reduce_token_size(openai_api_key, system_prompt, history, chatbot, all_token_counts, top_p, temperature, stream=False, hidden=True)
+        for chatbot, history, status_text, all_token_counts in iter:
             status_text = f"Token 达到上限，已自动降低Token计数至 {status_text}"
-            yield chatbot, history, status_text, token_count
+            yield chatbot, history, status_text, all_token_counts
 
 
-def retry(openai_api_key, system_prompt, history, chatbot, token_count, top_p, temperature, stream=False):
+def retry(openai_api_key, system_prompt, history, chatbot, token_count, top_p, temperature, stream=False, selected_model = MODELS[0]):
     print("重试中……")
     if len(history) == 0:
         yield chatbot, history, f"{standard_error_msg}上下文是空的", token_count
@@ -247,15 +263,15 @@ def retry(openai_api_key, system_prompt, history, chatbot, token_count, top_p, t
     history.pop()
     inputs = history.pop()["content"]
     token_count.pop()
-    iter = predict(openai_api_key, system_prompt, history, inputs, chatbot, token_count, top_p, temperature, stream=stream)
+    iter = predict(openai_api_key, system_prompt, history, inputs, chatbot, token_count, top_p, temperature, stream=stream, selected_model=selected_model)
     print("重试完毕")
     for x in iter:
         yield x
 
 
-def reduce_token_size(openai_api_key, system_prompt, history, chatbot, token_count, top_p, temperature, stream=False, hidden=False):
+def reduce_token_size(openai_api_key, system_prompt, history, chatbot, token_count, top_p, temperature, stream=False, hidden=False, selected_model = MODELS[0]):
     print("开始减少token数量……")
-    iter = predict(openai_api_key, system_prompt, history, summarize_prompt, chatbot, token_count, top_p, temperature, stream=stream, should_check_token_count=False)
+    iter = predict(openai_api_key, system_prompt, history, summarize_prompt, chatbot, token_count, top_p, temperature, stream=stream, selected_model = selected_model, should_check_token_count=False)
     for chatbot, history, status_text, previous_token_count in iter:
         history = history[-2:]
         token_count = previous_token_count[-1:]
@@ -265,7 +281,7 @@ def reduce_token_size(openai_api_key, system_prompt, history, chatbot, token_cou
     print("减少token数量完毕")
 
 
-def delete_last_conversation(chatbot, history, previous_token_count, streaming):
+def delete_last_conversation(chatbot, history, previous_token_count):
     if len(chatbot) > 0 and standard_error_msg in chatbot[-1][1]:
         print("由于包含报错信息，只删除chatbot记录")
         chatbot.pop()
@@ -280,7 +296,7 @@ def delete_last_conversation(chatbot, history, previous_token_count, streaming):
     if len(previous_token_count) > 0:
         print("删除了一组对话的token计数记录")
         previous_token_count.pop()
-    return chatbot, history, previous_token_count, construct_token_message(sum(previous_token_count), streaming)
+    return chatbot, history, previous_token_count, construct_token_message(sum(previous_token_count))
 
 
 def save_chat_history(filename, system, history, chatbot):