Merge pull request #16 from codedog-ai/feat/#2-coding-problem-mgmt

README.md

@@ -19,21 +19,25 @@ license: mit
 
 ## config
 
-| Environment Variable          | Necessary | Default | Description                              |
-| ----------------------------- | --------- | ------- | ---------------------------------------- |
-| CODEDOG_SERVER                | No        | 0.0.0.0 | Server address                           |
-| CODEDOG_PORT                  | No        | 32167   | Server port                              |
-| CODEDOG_WORKER_NUM            | No        | 1       | Server worker number                     |
-| OPENAI_API_KEY                | Yes       |         | Api Key for calling openai api           |
-| OPENAI_PROXY                  | No        |         | openai proxy                             |
-| AZURE_OPENAI                  | No        |         | use azure openai if not blank            |
-| AZURE_OPENAI_API_KEY          | No        |         | azure openai api key                     |
-| AZURE_OPENAI_API_BASE         | No        |         | azure openai api base                    |
-| AZURE_OPENAI_DEPLOYMENT_ID    | No        |         | azure openai deployment id for gpt 3.5   |
-| AZURE_OPENAI_EMBEDDING_DEP_ID | No        |         | azure openai deployment id for embedding |
-| QDRANT_API                    | No        |         | use qdrant api if not blank              |
-| QDRANT_API_KEY                | No        |         | qdrant api key                           |
-| QDRANT_API_BASE               | No        |         | qdrant api base                          |
+| Environment Variable          | Necessary | Default   | Description                              |
+| ----------------------------- | --------- | --------- | ---------------------------------------- |
+| CODEDOG_SERVER                | No        | 0.0.0.0   | Server address                           |
+| CODEDOG_PORT                  | No        | 32167     | Server port                              |
+| CODEDOG_WORKER_NUM            | No        | 1         | Server worker number                     |
+| OPENAI_API_KEY                | Yes       |           | Api Key for calling openai api           |
+| OPENAI_PROXY                  | No        |           | openai proxy                             |
+| AZURE_OPENAI                  | No        |           | use azure openai if not blank            |
+| AZURE_OPENAI_API_KEY          | No        |           | azure openai api key                     |
+| AZURE_OPENAI_API_BASE         | No        |           | azure openai api base                    |
+| AZURE_OPENAI_DEPLOYMENT_ID    | No        |           | azure openai deployment id for gpt 3.5   |
+| AZURE_OPENAI_EMBEDDING_DEP_ID | No        |           | azure openai deployment id for embedding |
+| QDRANT_API                    | No        |           | use qdrant api if not blank              |
+| QDRANT_API_KEY                | No        |           | qdrant api key                           |
+| QDRANT_API_BASE               | No        |           | qdrant api base                          |
+| REDIS_HOST                    | No        | 127.0.0.1 | redis host. used for orm                 |
+| REDIS_PORT                    | No        | 6379      | redis port                               |
+| REDIS_PASSWORD                | No        |           | redis password                           |
+| REDIS_SSL                     | No        |           | connect use ssl if not blank             |
 
 ## setup
 

edu_assistant/learning_tasks/coding_problem.py

@@ -4,9 +4,11 @@ from langchain.chains.base import Chain
 from langchain.memory import ConversationBufferMemory
 from langchain.schema import BaseRetriever
 from pydantic import BaseModel, Field
+from pydantic_redis import Model, Store
 
 from edu_assistant.learning_tasks.base import BaseTask
 from edu_assistant.utils.langchain_utils import load_llm
+from edu_assistant.utils.redis_utils import get_redis_config
 
 TEMPLATE = """The following is a friendly conversation between a human and an ai.
 The ai is talkative and provides lots of specific details from its context.
@@ -38,24 +40,33 @@ The code is written by a student aged 5-10 and mostly like to buggy or bad perfo
 DEFAULT_FIRST_QUESTION = "请问这段代码中有什么问题吗？"
 
 
-class CodingProblem(BaseModel):
+class CodingProblem(Model):
+    _primary_key_field: str = "title"
+    title: str = Field()
     question: str = Field()
     standard_answer: str = Field(default="")
     analysis: str = Field(default="")
-    extra: list[str] = Field(default_factory=lambda _: list())
+    language: str = Field(default="")
+    extra: str = Field(default_factory=lambda: list())
 
     # TODO: Add cache to expr function with pydantic 2 computed_field decorator.
     # Wait for langchain to support pydantic2.
 
+    @staticmethod
+    def enable_redis_orm():
+        store = Store(name="coding_problems", redis_config=get_redis_config(), life_span_in_seconds=3600 * 24 * 30)
+
+        store.register_model(CodingProblem)
+
     def expr(self, lang=""):
-        expr = f"Question:\n```\n{self.question}\n```\n"
+        expr = f"## Question\n\n```\n{self.question}\n```\n\n"
         expr += (
-            f"Question Standard Answer (There might be others):\n```{lang}\n{self.standard_answer}```"
+            f"## Standard Answer (There might be others)\n\n```{lang}\n{self.standard_answer}\n```\n\n"
             if self.standard_answer
             else ""
         )
-        expr += f"Solution Analysis:\n```{self.analysis}```\n" if self.analysis else ""
-        expr += "".join(self.extra) + "\n"
+        expr += f"## Analysis\n\n```\n{self.analysis}\n```\n\n" if self.analysis else ""
+        expr += "## Extra\n\n" + "".join(self.extra) + "\n"
         return expr
 
     def __str__(self):
@@ -84,6 +95,16 @@ class CodingProblemAnalysis(BaseTask):
         self._session_store = {}
         self._knowledge = knowledge
 
+    @staticmethod
+    def build_coding_problem(question: str, standard_answer: str = "", analysis: str = "", extra: list[str] = None):
+        extra = [] if extra is None else extra
+        return CodingProblem(question=question, standard_answer=standard_answer, analysis=analysis, extra=extra)
+
+    @staticmethod
+    def build_coding_answer(answer: str, extra: list[str] = None):
+        extra = [] if extra is None else extra
+        return CodingAnswer(answer=answer, extra=extra)
+
     def start_analysis(self, problem: CodingProblem, answer: CodingAnswer, first_question: str = None) -> dict:
         """start analysis of a coding problem and incorrect answer.
 
@@ -116,6 +137,9 @@ class CodingProblemAnalysis(BaseTask):
         """
         assert question
 
+        if session_id not in self._session_store:
+            return {}
+
         chain = self._session_store[session_id]
 
         result = chain({"input": question})
@@ -129,7 +153,9 @@ class CodingProblemAnalysis(BaseTask):
         memory = ConversationBufferMemory()
         prompt = PromptTemplate.from_template(
             TEMPLATE.format(
-                instruction=self.instruction, problem=problem.expr(lang=self.lang), answer=answer.expr(lang=self.lang)
+                instruction=self.instruction,
+                problem=problem.expr(lang=problem.language or self.lang or ""),
+                answer=answer.expr(lang=problem.language or self.lang or ""),
             )
         )
 
@@ -148,13 +174,3 @@ class CodingProblemAnalysis(BaseTask):
                 return_source_documents=True,
                 combine_docs_chain_kwargs={"prompt": prompt},
             )
-
-    @staticmethod
-    def build_coding_problem(question: str, standard_answer: str = "", analysis: str = "", extra: list[str] = None):
-        extra = [] if extra is None else extra
-        return CodingProblem(question=question, standard_answer=standard_answer, analysis=analysis, extra=extra)
-
-    @staticmethod
-    def build_coding_answer(answer: str, extra: list[str] = None):
-        extra = [] if extra is None else extra
-        return CodingAnswer(answer=answer, extra=extra)

edu_assistant/utils/redis_utils.py

@@ -0,0 +1,20 @@
+import os
+from functools import lru_cache
+
+from pydantic_redis import RedisConfig
+from redis import Redis
+
+host = os.environ.get("REDIS_HOST", "localhost")
+port = int(os.environ.get("REDIS_PORT", 6379))
+password = os.environ.get("REDIS_PASSWORD", None)
+ssl = True if os.environ.get("REDIS_SSL") else False
+
+
+@lru_cache(maxsize=1)
+def get_redis_client():
+    return Redis(host=host, port=port, password=password, ssl=ssl)
+
+
+@lru_cache(maxsize=1)
+def get_redis_config():
+    return RedisConfig(host=host, port=port, password=password, ssl=ssl)

examples/coding_problem.py

@@ -4,9 +4,11 @@ from edu_assistant.learning_tasks import CodingProblemAnalysis
 
 # 创建一个CodingProblem实例
 problem = CodingProblemAnalysis.build_coding_problem(
+    title="求解最大值",
     question="请编写一个函数，该函数接收一个整数列表，并返回该列表的最大值。",
     standard_answer="def find_max(lst):\n\treturn max(lst)",
     analysis="在这个问题中，我们需要使用Python的内置函数max来找到列表的最大值。",
+    language="python",
 )
 
 # 创建一个CodingAnswer实例

examples/coding_problem_orm.py

@@ -0,0 +1,21 @@
+import json
+
+from edu_assistant.learning_tasks.coding_problem import CodingProblem
+
+
+def load_problems(file_path: str):
+    """load problems from json file and insert into redis orm.
+
+    Args:
+        file_path (str): file path to json file.
+    """
+    with open(file_path, "r", encoding="utf-8") as f:
+        problems_data = json.load(f)
+
+    CodingProblem.enable_redis_orm()
+    problems = [CodingProblem.parse_obj(problem_data) for problem_data in problems_data]
+    CodingProblem.insert(problems)
+
+
+if __name__ == "__main__":
+    load_problems("examples/coding_problems.json")

examples/coding_problems.json

@@ -0,0 +1,9 @@
+[
+    {
+        "title": "Find Max Number",
+        "question": "请编写一个函数，该函数接收一个整数列表，并返回该列表的最大值",
+        "standard_answer": "def find_max(lst):\n\treturn max(lst)",
+        "analysis": "在这个问题中，我们需要使用Python的内置函数max来找到列表的最大值。",
+        "language": "python"
+    }
+]

pyproject.toml

@@ -18,6 +18,8 @@ uvicorn = "^0.21.1"
 qdrant-client = "^1.3.1"
 tiktoken = "^0.4.0"
 gradio = "^3.37.0"
+redis = "^4.6.0"
+pydantic-redis = "^0.4.3"
 
 [tool.poetry.group.dev]
 optional = true

webui/coding_problem.py

@@ -1,4 +1,197 @@
+import json
+
 import gradio as gr
+from fastapi.encoders import jsonable_encoder
+from langchain.callbacks import get_openai_callback
+
+from edu_assistant.learning_tasks.coding_problem import (
+    DEFAULT_FIRST_QUESTION,
+    CodingProblem,
+    CodingProblemAnalysis,
+)
+
+CodingProblem.enable_redis_orm()
+task = CodingProblemAnalysis()
+
+
+def get_problems() -> list[str]:
+    data = CodingProblem.select(columns=["title"])
+    if not data:
+        return []
+    titles = [problem_data["title"] for problem_data in data]
+    return titles
+
+
+def update_problems():
+    titles = get_problems()
+    gr.Info("更新题目列表成功")
+    return gr.Dropdown.update(choices=titles)
+
+
+def select_problem(title: str):
+    problem: CodingProblem = CodingProblem.select(ids=[title])[0]
+    return (
+        problem.expr(),
+        problem.title,
+        problem.language,
+        problem.question,
+        problem.analysis,
+        problem.standard_answer,
+        json.dumps(problem.extra, ensure_ascii=False, indent=4),
+    )
+
+
+def update_problem(title, language, problem, analysis, answer, extra):
+    # TODO: add language
+    try:
+        extra_data = json.loads(extra)
+    except json.JSONDecodeError:
+        extra_data = [extra]
+
+    CodingProblem.update(
+        title,
+        data={
+            "title": title,
+            "language": language,
+            "question": problem,
+            "analysis": analysis,
+            "standard_answer": answer,
+            "extra": extra_data,
+        },
+    )
+    gr.Info("更新题目成功")
+
+
+def delete_problem(title):
+    CodingProblem.delete(ids=[title])
+    gr.Info("删除题目成功")
+
+    return "", "", "", "", "", "", "", ""
+
+
+def analysis_problem(title, code, extra: str = ""):
+    problem = CodingProblem.select(ids=[title])[0]
+    answer = CodingProblemAnalysis.build_coding_answer(answer=code)
+
+    with get_openai_callback() as cb:
+        result = task.start_analysis(problem, answer)
+        status = {"tokens": cb.total_tokens, "cost": f"${cb.total_cost:.4f}"}
+
+    answer = result["response"]
+    session_id = result["session_id"]
+    docs = jsonable_encoder(result.get("source_documents", []))
+    return [(DEFAULT_FIRST_QUESTION, answer)], session_id, status, docs
+
+
+def chat(message, chat_history, session_id):
+    if not session_id:
+        return "", "", {"tokens": 0}, []
+    with get_openai_callback() as cb:
+        result = task.ask(message, session_id=session_id)
+        if not result:
+            raise gr.Error("Session expired. Please recreate a new problem analysis session.")
+
+        session_id = result["session_id"]
+        docs = jsonable_encoder(result.get("source_documents", []))
+
+        bot_message = result["response"]
+        chat_history.append((message, bot_message))
+
+        status = {"tokens": cb.total_tokens, "cost": f"${cb.total_cost:.4f}"}
+
+    return "", chat_history, status, docs
+
 
 with gr.Blocks() as coding_problem_ui:
-    pass
+    with gr.Row():
+        with gr.Column(scale=6):
+            problem_selector = gr.Dropdown(choices=get_problems(), show_label=False, interactive=True)
+        with gr.Column():
+            refresh_btn = gr.Button(value="刷新")
+    with gr.Row():
+        with gr.Column(scale=6):
+            with gr.Tab(label="错误代码分析"):
+                with gr.Row():
+                    with gr.Column(scale=3):
+                        with gr.Row():
+                            problem_view = gr.Markdown(label="题目")
+                        with gr.Row():
+                            code_view = gr.Textbox(label="代码", lines=10, interactive=True)
+                    with gr.Column(scale=3):
+                        with gr.Row():
+                            chat_box = gr.Chatbot(height=500)
+                        with gr.Row():
+                            chat_input = gr.Textbox(interactive=True)
+                    with gr.Column():
+                        with gr.Row():
+                            analysis_btn = gr.Button(value="分析")
+                        with gr.Row():
+                            status = gr.JSON(value="""{"tokens":0}""")
+                        with gr.Row():
+                            session_id = gr.Textbox(label="Session", interactive=False, value="")
+                        with gr.Row():
+                            docs = gr.JSON(value="""["docs"]""", label="Docs")
+                        with gr.Row():
+                            clear = gr.ClearButton([problem_view, code_view, problem_selector, session_id, docs])
+
+            with gr.Tab(label="题库管理"):
+                with gr.Row():
+                    with gr.Column(scale=6):
+                        with gr.Row():
+                            title_edit = gr.Textbox(label="标题", interactive=True)
+                        with gr.Row():
+                            language_edit = gr.Dropdown(
+                                choices=["python", "cpp", "java"],
+                                label="���言",
+                                interactive=True,
+                                allow_custom_value=True,
+                            )
+                    with gr.Column():
+                        manage_update = gr.Button(value="更新")
+                        manage_delete = gr.Button(value="删除", variant="stop")
+                with gr.Row():
+                    with gr.Column():
+                        problem_edit = gr.Textbox(label="题目", lines=10, max_lines=100, interactive=True)
+                    with gr.Column():
+                        analysis_edit = gr.Textbox(label="解析", lines=10, max_lines=100, interactive=True)
+                with gr.Row():
+                    answer_edit = gr.Textbox(label="标准答案", lines=10, max_lines=100, interactive=True)
+                with gr.Row():
+                    extra_edit = gr.Textbox(label="额外信息", lines=10, max_lines=100, interactive=True)
+
+    refresh_btn.click(update_problems, [], [problem_selector])
+    problem_selector.select(
+        select_problem,
+        [
+            problem_selector,
+        ],
+        [problem_view, title_edit, language_edit, problem_edit, analysis_edit, answer_edit, extra_edit],
+    )
+    analysis_btn.click(
+        analysis_problem,
+        [problem_selector, code_view],
+        [chat_box, session_id, status, docs],
+    )
+    chat_input.submit(chat, [chat_input, chat_box, session_id], [chat_input, chat_box, status, docs])
+
+    manage_update.click(
+        update_problem, [title_edit, language_edit, problem_edit, analysis_edit, answer_edit, extra_edit], []
+    )
+    manage_delete.click(
+        delete_problem,
+        [problem_selector],
+        [
+            problem_selector,
+            problem_view,
+            title_edit,
+            language_edit,
+            problem_edit,
+            analysis_edit,
+            answer_edit,
+            extra_edit,
+        ],
+    )
+
+if __name__ == "__main__":
+    coding_problem_ui.queue()
+    coding_problem_ui.launch(max_threads=2)

webui/qa.py

@@ -38,7 +38,7 @@ def recreate(instruction):
 
 
 def clear(msg, chatbot, session_id, telemetry, docs):
-    return "", "", '',  '{"tokens":0}', '["docs"]'
+    return "", "", "", '{"tokens":0}', '["docs"]'
 
 
 with gr.Blocks() as qa_ui:
@@ -52,7 +52,7 @@ with gr.Blocks() as qa_ui:
     with gr.Row():
         with gr.Column(scale=6):
             with gr.Row():
-                chatbot = gr.Chatbot(height=400)
+                chatbot = gr.Chatbot(height=500)
             with gr.Row():
                 msg = gr.Textbox()
         with gr.Column(scale=1):
@@ -62,7 +62,8 @@ with gr.Blocks() as qa_ui:
                 telemetry = gr.JSON(value="""{"tokens":0}""", label="Telemetry")
             with gr.Row():
                 docs = gr.JSON(value="""["docs"]""", label="Docs")
+            with gr.Row():
+                instruction = gr.Textbox(label="Instruction", value=DEFAULT_INSTRUCTION, interactive=False)
 
     clear_btn.click(clear, [msg, chatbot, session_id, telemetry, docs], [msg, chatbot, session_id, telemetry, docs])
     msg.submit(respond, [msg, chatbot, session_id], [msg, chatbot, session_id, telemetry, docs])
-    apply.click(recreate, [instruction], [])

webui/ui.py

@@ -1,8 +1,8 @@
 import gradio as gr
-from coding_problem import coding_problem_ui
-from qa import qa_ui
 
 from edu_assistant import version
+from webui.coding_problem import coding_problem_ui
+from webui.qa import qa_ui
 
 with gr.Blocks() as ui:
     with gr.Row():
@@ -15,4 +15,5 @@ with gr.Blocks() as ui:
         coding_problem_ui.render()
 
 if __name__ == "__main__":
+    ui.queue()
     ui.launch()