add feed back chat

.gitignore

@@ -0,0 +1 @@
+__pycache__/*

app.py

@@ -1,40 +1,75 @@
 import gradio as gr
-import openai
-import os
-from langchain.chains import ConversationChain
-from langchain.llms import OpenAI
-from langchain.memory import ConversationEntityMemory
-from langchain.memory.prompt import ENTITY_MEMORY_CONVERSATION_TEMPLATE
+from langchain.chains import LLMChain
+from langchain.chat_models import ChatOpenAI
+from langchain.document_loaders import TextLoader
+from memories import HumenFeedbackBufferMemory
+from promopts import FEEDBACK, FEEDBACK_PROMPT
 
-openai.api_key = os.environ.get("OPENAI_API_KEY")
+llm = ChatOpenAI(temperature=0.1)
 
-llm = OpenAI(model_name="gpt-3.5-turbo", stop="\n\n",
-             max_tokens=2048, temperature=0.5)
-entityMemory = ConversationEntityMemory(llm=llm)
+memory = HumenFeedbackBufferMemory(
+    input_key="input", human_prefix="Answer", ai_prefix="AI")
 
-conversation = ConversationChain(
-    llm=llm, prompt=ENTITY_MEMORY_CONVERSATION_TEMPLATE, verbose=True, memory=entityMemory)
+llmchain = LLMChain(
+    llm=llm,
+    prompt=FEEDBACK_PROMPT,
+    memory=memory,
+    verbose=True
+)
 
+"""读取document/business_context.py文件内容作为context"""
+context_path = "./documents/business_context.md"
+textloader = TextLoader(context_path)
+CONTEXT = textloader.load()[0].page_content
 
-def answer(question, history=[]):
-    history.append(question)
-    response = conversation.predict(input=question)
-    history.append(response)
-    responses = [(u, b) for u, b in zip(history[::2], history[1::2])]
-    return responses, history
 
+def sendMessage(chatbot, input):
+    chatbot.append((
+        (None if len(input) == 0 else input), None))
+    return chatbot
 
-with gr.Blocks(css="#chatbot{height:300px} .overflow-y-auto{height:500px}") as demo:
-    chatbot = gr.Chatbot(elem_id="chatbot")
-    state = gr.State([])
 
-    with gr.Row():
-        txt = gr.Textbox(show_label=False, placeholder="Enter text and press enter").style(
-            container=False)
+def clearMemory(chatbot):
+    chatbot.clear()
+    memory.clear()
+    return chatbot, ""
+
+
+def feedBack(context, story, chatbot=[], input=""):
+    if len(input) > 0:
+        context += (f"\n- {input}")
+        with open(context_path, 'w') as f:
+            f.write(context)
+
+    response = llmchain.run(
+        input=(input if len(input) == 0 else ("Answer: " + input)), context=context, story=story, stop="\nAnswer:")
+    chatbot[-1][1] = response
+    return chatbot, "", context
 
+
+with gr.Blocks() as demo:
     with gr.Row():
-        button = gr.Button("Submit")
+        with gr.Column():
+            chatbot = gr.Chatbot(elem_id="chatbot").style()
+
+            with gr.Row():
+                txt = gr.Textbox(show_label=False, placeholder="Enter text and press enter").style(
+                    container=False)
+        with gr.Column():
+            with gr.Row():
+                context = gr.Textbox(show_label=True, label="Context", value=CONTEXT, placeholder="Enter Context").style(
+                    container=False)
+
+            with gr.Row():
+                story = gr.Textbox(show_label=True, label="Story", placeholder="Enter Story").style(
+                    container=False)
+
+            with gr.Row():
+                button = gr.Button("Generate Scenarios")
 
-    button.click(answer, [txt, state], [chatbot, state])
+    button.click(clearMemory, [chatbot], [chatbot, txt]).then(sendMessage, [chatbot, txt], [chatbot]).then(
+        feedBack, [context, story, chatbot], [chatbot, txt])
+    txt.submit(sendMessage, [chatbot, txt], [chatbot]).then(
+        feedBack, [context, story, chatbot, txt], [chatbot, txt, context])
 
 demo.launch()

documents/business_context.md

@@ -0,0 +1,13 @@
+ABstract是一个AB测试平台，整合了配置管理和实验管理，为企业提供数据驱动的决策能力。例如，一个用户想要创建一个实验来测试两个不同版本的网页，以了解哪个版本能带来更多的服务注册。他们将创建一个实验，包含两个分组（bucket），每个分组对应一个网页版本，并将不同的用户分配到每个分组以比较结果。
+
+每个实验可以包含1个或多个分组，每个分组都有一个不同的配置。实验信息包括：实验名称、实验描述、开始和结束时间、抽样百分比、识别实验的关键字、每个分组的名称及其关联的配置和分配给每个分组的流量百分比，以及将用户分配到分组的目标规则。
+
+实验和分组可以一起配置。实验中的配置用于为用户提供不同的交互体验。一个实验最多可以创建10个分组。分组配置将作为分组的一个配置项与分组关联。
+
+实验分为灰度发布、AB对比和多分组实验三种类型。一个实验应该包含：实验名称、实验描述、开始和结束时间、抽样百分比、识别实验的关键字、每个分组的名称及其关联的配置和分配给每个分组的流量百分比，以及将用户分配到分组的目标规则。
+
+实验的目的是将用户随机分配到不同的分组，不同分组中的用户获得不同的软件交互体验。通过收集不同分组中用户的行为数据，对比不同设计的效果。一个实验至少包含1个，最多10个分组。
+
+用户在同一个编辑界面编辑实验和分组信息，点击保存按钮时一起提交。创建实验时应该包含：实验名称、分流比例、开始时间、截止时间、分组（名称、关键字、分流比例）。实验类型分为灰度发布、AB对比和多分组实验三种。
+- 实验的目标是通过将用户随机分组，收集不同组用户的行为数据，对比分析不同设计的效果，找出最有设计
+- 实验有三种： 1.灰度发布（一个实验只有一个分组，通过实验采样比例控制新特性的发布比例）2. A/B对比（一个实验有两个分组，在用户总量中采样部分用户进行实验，对比新旧两组的数据）3. 多分组实验（一个实验有多个分组，在用户总量中采样部分用户进行实验，对比不同设计的效果））

memories.py

@@ -0,0 +1,61 @@
+from typing import Any, Dict, List, Optional
+from langchain.memory.chat_memory import BaseChatMemory
+from langchain.schema import BaseMessage, HumanMessage, AIMessage, SystemMessage, ChatMessage
+
+
+def get_buffer_string(
+    messages: List[BaseMessage], human_prefix: str = "Human", ai_prefix: str = "AI"
+) -> str:
+    """Get buffer string of messages."""
+    string_messages = []
+    for m in messages:
+        if isinstance(m, HumanMessage):
+            print("HumanMessage: " + m.content)
+            role = human_prefix + ":"
+        elif isinstance(m, AIMessage):
+            print("AIMessage" + m.content)
+            role = ""
+        elif isinstance(m, SystemMessage):
+            print("SystemMessage")
+            role = "System:"
+        elif isinstance(m, ChatMessage):
+            print("ChatMessage")
+            role = m.role + ":"
+        else:
+            raise ValueError(f"Got unsupported message type: {m}")
+
+        string_messages.append(f"{m.content}")
+
+    return "\n".join(string_messages)
+
+
+class HumenFeedbackBufferMemory(BaseChatMemory):
+    """Buffer for storing conversation memory."""
+
+    human_prefix: str = "Human"
+    ai_prefix: str = "AI"
+    memory_key: str = "history"  #: :meta private:
+
+    @property
+    def buffer(self) -> Any:
+        """String buffer of memory."""
+        if self.return_messages:
+            return self.chat_memory.messages
+        else:
+            return get_buffer_string(
+                self.chat_memory.messages,
+                human_prefix=self.human_prefix,
+                ai_prefix=self.ai_prefix,
+            )
+
+    @property
+    def memory_variables(self) -> List[str]:
+        """Will always return list of memory variables.
+
+        :meta private:
+        """
+        return [self.memory_key]
+
+    def load_memory_variables(self, inputs: Dict[str, Any]) -> Dict[str, Any]:
+        """Return history buffer."""
+        return {self.memory_key: self.buffer}

promopts.py

@@ -0,0 +1,28 @@
+from langchain.prompts import PromptTemplate
+
+FEEDBACK = """You are a business analyst who is familiar with specification by example. I'm the domain expert.
+
+===CONTEXT
+{context}
+===END OF CONTXT
+
+===USER STORY
+{story}
+===END OF USER STORY
+
+Explain the user story as scenarios. use the following format:
+
+Thought: you should always think about what is still uncertain about the user story. Ignore technical concerns.
+Question: the Question to ask to clarify the user story
+Answer: the answer I responded to the question
+... (this Thought/Question/Answer repeat at least 3 times, at most 10 times)
+Thought: I know enough to explain the user story
+Scenarios: List all possible scenarios with concrete example in Given/When/Then style
+
+
+{history}
+{input}"""
+
+FEEDBACK_PROMPT = PromptTemplate(
+    input_variables=["context", "story", "history", "input"], template=FEEDBACK,
+)