大幅度改进代码质量

ChuanhuChatbot.py

@@ -8,9 +8,6 @@ from presets import *
 
 
 my_api_key = ""    # 在这里输入你的 API 密钥
-HIDE_MY_KEY = False # 如果你想在UI中隐藏你的 API 密钥，将此值设置为 True
-
-gr.Chatbot.postprocess = postprocess
 
 #if we are running in Docker
 if os.environ.get('dockerrun') == 'yes':
@@ -42,12 +39,17 @@ else:
             if username != "" and password != "":
                 authflag = True
 
+gr.Chatbot.postprocess = postprocess
+
 with gr.Blocks(css=customCSS) as demo:
     gr.HTML(title)
-    keyTxt = gr.Textbox(show_label=True, placeholder=f"在这里输入你的OpenAI API-key...",
-                        value=my_api_key, label="API Key", type="password", visible=not HIDE_MY_KEY).style(container=True)
+    with gr.Row():
+        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)
+        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))
     TRUECOMSTANT = gr.State(True)
     FALSECONSTANT = gr.State(False)
@@ -55,16 +57,16 @@ with gr.Blocks(css=customCSS) as demo:
 
     with gr.Row():
         with gr.Column(scale=12):
-            txt = gr.Textbox(show_label=False, placeholder="在这里输入").style(
+            user_input = gr.Textbox(show_label=False, placeholder="在这里输入").style(
                 container=False)
         with gr.Column(min_width=50, scale=1):
             submitBtn = gr.Button("🚀", variant="primary")
     with gr.Row():
         emptyBtn = gr.Button("🧹 新的对话")
         retryBtn = gr.Button("🔄 重新生成")
-        delLastBtn = gr.Button("🗑️ 删除上条对话")
+        delLastBtn = gr.Button("🗑️ 删除最近一条对话")
         reduceTokenBtn = gr.Button("♻️ 总结对话")
-    statusDisplay = gr.Markdown("status: ready")
+    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)
     with gr.Accordion(label="加载Prompt模板", open=False):
@@ -105,26 +107,34 @@ with gr.Blocks(css=customCSS) as demo:
     gr.Markdown(description)
 
 
-    txt.submit(predict, [txt, top_p, temperature, keyTxt,
-               chatbot, history, systemPromptTxt], [chatbot, history, statusDisplay])
-    txt.submit(reset_textbox, [], [txt])
-    submitBtn.click(predict, [txt, top_p, temperature, keyTxt, chatbot,
-                    history, systemPromptTxt], [chatbot, history, statusDisplay], show_progress=True)
-    submitBtn.click(reset_textbox, [], [txt])
-    emptyBtn.click(reset_state, outputs=[chatbot, history])
-    retryBtn.click(predict, [txt, top_p, temperature, keyTxt, chatbot, history,
-                   systemPromptTxt, TRUECOMSTANT], [chatbot, history, statusDisplay], show_progress=True)
-    delLastBtn.click(delete_last_conversation, [chatbot, history], [
-                     chatbot, history], show_progress=True)
-    reduceTokenBtn.click(predict, [txt, top_p, temperature, keyTxt, chatbot, history,
-                         systemPromptTxt, FALSECONSTANT, TRUECOMSTANT], [chatbot, history, statusDisplay], show_progress=True)
+    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(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(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)
+
+    delLastBtn.click(delete_last_conversation, [chatbot, history, token_count, use_streaming_checkbox], [
+                     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)
+
     saveHistoryBtn.click(save_chat_history, [
                   saveFileName, systemPromptTxt, history, chatbot], None, show_progress=True)
+
     saveHistoryBtn.click(get_history_names, None, [historyFileSelectDropdown])
+
     historyRefreshBtn.click(get_history_names, None, [historyFileSelectDropdown])
+
     historyReadBtn.click(load_chat_history, [historyFileSelectDropdown, systemPromptTxt, history, chatbot],  [saveFileName, systemPromptTxt, history, chatbot], show_progress=True)
+
     templateRefreshBtn.click(get_template_names, None, [templateFileSelectDropdown])
+
     templaeFileReadBtn.click(load_template, [templateFileSelectDropdown],  [promptTemplates, templateSelectDropdown], show_progress=True)
+
     templateApplyBtn.click(get_template_content, [promptTemplates, templateSelectDropdown, systemPromptTxt],  [systemPromptTxt], show_progress=True)
 
 print("川虎的温馨提示：访问 http://localhost:7860 查看界面")

presets.py

@@ -29,3 +29,13 @@ pre code {
     box-shadow: inset 0px 8px 16px hsla(0, 0%, 0%, .2)
 }
 """
+
+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
+max_token_streaming = 3000 # 流式对话时的最大 token 数
+timeout_streaming = 5 # 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
+timeout_all = 200 # 非流式对话时的超时时间
+enable_streaming_option = False  # 是否启用选择选择是否实时显示回答的勾选框
+HIDE_MY_KEY = False # 如果你想在UI中隐藏你的 API 密钥，将此值设置为 True

utils.py

@@ -14,6 +14,7 @@ import requests
 import csv
 import mdtex2html
 from pypinyin import lazy_pinyin
+from presets import *
 
 if TYPE_CHECKING:
     from typing import TypedDict
@@ -51,7 +52,6 @@ def parse_text(text):
     lines = text.split("\n")
     lines = [line for line in lines if line != ""]
     count = 0
-    firstline = False
     for i, line in enumerate(lines):
         if "```" in line:
             count += 1
@@ -79,54 +79,33 @@ def parse_text(text):
     text = "".join(lines)
     return text
 
-def predict(inputs, top_p, temperature, openai_api_key, chatbot=[], history=[], system_prompt=initial_prompt, retry=False, summary=False, retry_on_crash = False, stream = True):  # repetition_penalty, top_k
+def construct_text(role, text):
+    return {"role": role, "content": text}
 
-    if retry_on_crash:
-        retry = True
+def construct_user(text):
+    return construct_text("user", text)
 
+def construct_system(text):
+    return construct_text("system", text)
+
+def construct_assistant(text):
+    return construct_text("assistant", text)
+
+def construct_token_message(token, stream=False):
+    extra = "【仅包含回答的计数】 " if stream else ""
+    return f"{extra}Token 计数: {token}"
+
+def get_response(openai_api_key, system_prompt, history, temperature, top_p, stream):
     headers = {
         "Content-Type": "application/json",
         "Authorization": f"Bearer {openai_api_key}"
     }
 
-    chat_counter = len(history) // 2
-
-    print(f"chat_counter - {chat_counter}")
-
-    messages = []
-    if chat_counter:
-        for index in range(0, 2*chat_counter, 2):
-            temp1 = {}
-            temp1["role"] = "user"
-            temp1["content"] = history[index]
-            temp2 = {}
-            temp2["role"] = "assistant"
-            temp2["content"] = history[index+1]
-            if temp1["content"] != "":
-                if temp2["content"] != "" or retry:
-                    messages.append(temp1)
-                    messages.append(temp2)
-            else:
-                messages[-1]['content'] = temp2['content']
-    if retry and chat_counter:
-        if retry_on_crash:
-            messages = messages[-6:]
-        messages.pop()
-    elif summary:
-        history = [*[i["content"] for i in messages[-2:]], "我们刚刚聊了什么？"]
-        messages.append(compose_user(
-            "请帮我总结一下上述对话的内容，实现减少字数的同时，保证对话的质量。在总结中不要加入这一句话。"))
-    else:
-        temp3 = {}
-        temp3["role"] = "user"
-        temp3["content"] = inputs
-        messages.append(temp3)
-        chat_counter += 1
-    messages = [compose_system(system_prompt), *messages]
-    # messages
+    history = [construct_system(system_prompt), *history]
+
     payload = {
         "model": "gpt-3.5-turbo",
-        "messages": messages,  # [{"role": "user", "content": f"{inputs}"}],
+        "messages": history,  # [{"role": "user", "content": f"{inputs}"}],
         "temperature": temperature,  # 1.0,
         "top_p": top_p,  # 1.0,
         "n": 1,
@@ -134,94 +113,129 @@ def predict(inputs, top_p, temperature, openai_api_key, chatbot=[], history=[],
         "presence_penalty": 0,
         "frequency_penalty": 0,
     }
-
-    if not summary:
-        history.append(inputs)
+    if stream:
+        timeout = timeout_streaming
     else:
-        print("精简中...")
+        timeout = timeout_all
+    response = requests.post(API_URL, headers=headers, json=payload, stream=True, timeout=timeout)
+    return response
 
-    print(f"payload: {payload}")
-    # make a POST request to the API endpoint using the requests.post method, passing in stream=True
+def stream_predict(openai_api_key, system_prompt, history, inputs, chatbot, previous_token_count, top_p, temperature):
+    def get_return_value():
+        return chatbot, history, status_text, [*previous_token_count, token_counter]
+    token_counter = 0
+    partial_words = ""
+    counter = 0
+    status_text = "OK"
+    history.append(construct_user(inputs))
     try:
-        response = requests.post(API_URL, headers=headers, json=payload, stream=True)
-    except:
-        history.append("")
-        chatbot.append((inputs, ""))
-        yield history, chatbot, f"获取请求失败，请检查网络连接。"
+        response = get_response(openai_api_key, system_prompt, history, temperature, top_p, True)
+    except requests.exceptions.ConnectTimeout:
+        status_text = standard_error_msg + error_retrieve_prompt
+        yield get_return_value()
         return
 
-    token_counter = 0
-    partial_words = ""
+    chatbot.append((parse_text(inputs), ""))
+    yield get_return_value()
 
-    counter = 0
-    if stream:
-        chatbot.append((parse_text(history[-1]), ""))
-        for chunk in response.iter_lines():
-            if counter == 0:
-                counter += 1
-                continue
+    for chunk in response.iter_lines():
+        if counter == 0:
             counter += 1
-            # check whether each line is non-empty
-            if chunk:
-                # decode each line as response data is in bytes
-                try:
-                    if len(json.loads(chunk.decode()[6:])['choices'][0]["delta"]) == 0:
-                        chunkjson = json.loads(chunk.decode()[6:])
-                        status_text = f"id: {chunkjson['id']}, finish_reason: {chunkjson['choices'][0]['finish_reason']}"
-                        yield chatbot, history, status_text
-                        break
-                except Exception as e:
-                    if not retry_on_crash:
-                        print("正在尝试使用缩短的context重新生成……")
-                        chatbot.pop()
-                        history.append("")
-                        yield next(predict(inputs, top_p, temperature, openai_api_key, chatbot, history, system_prompt, retry, summary=False, retry_on_crash=True, stream=False))
-                    else:
-                        msg = "☹️发生了错误：生成失败，请检查网络"
-                        print(msg)
-                        history.append(inputs, "")
-                        chatbot.append(inputs, msg)
-                        yield chatbot, history, "status: ERROR"
+            continue
+        counter += 1
+        # check whether each line is non-empty
+        if chunk:
+            chunk = chunk.decode()
+            chunklength = len(chunk)
+            chunk = json.loads(chunk[6:])
+            # 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, stream=True)
+                if finish_reason == "stop":
+                    yield get_return_value()
                     break
-                chunkjson = json.loads(chunk.decode()[6:])
-                status_text = f"id: {chunkjson['id']}, finish_reason: {chunkjson['choices'][0]['finish_reason']}"
-                partial_words = partial_words + \
-                    json.loads(chunk.decode()[6:])[
-                        'choices'][0]["delta"]["content"]
+                partial_words = partial_words + chunk['choices'][0]["delta"]["content"]
                 if token_counter == 0:
-                    history.append(" " + partial_words)
+                    history.append(construct_assistant(" " + partial_words))
                 else:
-                    history[-1] = partial_words
-                chatbot[-1] = (parse_text(history[-2]), parse_text(history[-1]))
+                    history[-1] = construct_assistant(partial_words)
+                chatbot[-1] = (parse_text(inputs), parse_text(partial_words))
                 token_counter += 1
-                yield chatbot, history, status_text
-    else:
-        try:
-            responsejson = json.loads(response.text)
-            content = responsejson["choices"][0]["message"]["content"]
-            history.append(content)
-            chatbot.append((parse_text(history[-2]), parse_text(content)))
-            status_text = "精简完成"
-        except:
-            chatbot.append((parse_text(history[-1]), "☹️发生了错误，请检查网络连接或者稍后再试。"))
-            status_text = "status: ERROR"
-        yield chatbot, history, status_text
+                yield get_return_value()
 
 
-
-def delete_last_conversation(chatbot, history):
+def predict_all(openai_api_key, system_prompt, history, inputs, chatbot, previous_token_count, top_p, temperature):
+    history.append(construct_user(inputs))
     try:
-        if "☹️发生了错误" in chatbot[-1][1]:
+        response = get_response(openai_api_key, system_prompt, history, temperature, top_p, False)
+    except requests.exceptions.ConnectTimeout:
+        status_text = standard_error_msg + error_retrieve_prompt
+        return chatbot, history, status_text, previous_token_count
+    response = json.loads(response.text)
+    content = response["choices"][0]["message"]["content"]
+    history.append(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))
+    status_text = construct_token_message(total_token_count)
+    return chatbot, history, status_text, previous_token_count
+
+
+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
+    if stream:
+        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
+    else:
+        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
+    if stream:
+        max_token = max_token_streaming
+    else:
+        max_token = max_token_all
+    if sum(token_count) > max_token and should_check_token_count:
+        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:
+            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
+
+
+def retry(openai_api_key, system_prompt, history, chatbot, token_count, top_p, temperature, stream=False):
+    if len(history) == 0:
+        yield chatbot, history, f"{standard_error_msg}上下文是空的", token_count
+        return
+    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)
+    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):
+    iter = predict(openai_api_key, system_prompt, history, summarize_prompt, chatbot, token_count, top_p, temperature, stream=stream, should_check_token_count=False)
+    for chatbot, history, status_text, previous_token_count in iter:
+        history = history[-2:]
+        token_count = previous_token_count[-1:]
+        if hidden:
             chatbot.pop()
-            print(history)
-            return chatbot, history
+        yield chatbot, history, construct_token_message(sum(token_count), stream=stream), token_count
+
+
+def delete_last_conversation(chatbot, history, previous_token_count, streaming):
+    if len(chatbot) > 0 and standard_error_msg in chatbot[-1][1]:
+        chatbot.pop()
+        return chatbot, history
+    if len(history) > 0:
         history.pop()
         history.pop()
+    if len(chatbot) > 0:
         chatbot.pop()
-        print(history)
-        return chatbot, history
-    except:
-        return chatbot, history
+    if len(previous_token_count) > 0:
+        previous_token_count.pop()
+    return chatbot, history, previous_token_count, construct_token_message(sum(previous_token_count), streaming)
+
 
 def save_chat_history(filename, system, history, chatbot):
     if filename == "":
@@ -298,7 +312,7 @@ def get_template_content(templates, selection, original_system_prompt):
         return original_system_prompt
 
 def reset_state():
-    return [], []
+    return [], [], [], construct_token_message(0)
 
 def compose_system(system_prompt):
     return {"role": "system", "content": system_prompt}