updated question guard

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+question_undetector.pkl filter=lfs diff=lfs merge=lfs -text

app.py

@@ -1,4 +1,5 @@
 import os
+import pickle
 import gradio as gr
 from openai import OpenAI
 
@@ -7,6 +8,16 @@ client = OpenAI(
     base_url="https://yo4x63mj3sbmgpwc.us-east-1.aws.endpoints.huggingface.cloud/v1/",
     api_key=os.environ.get("hf_token"),
 )
+with open("./question_undetector.pkl", "rb") as f:
+    (vectorizer, model) = pickle.load(f)
+
+
+def guard_question(question):
+    pred = model.predict(vectorizer.transform([question]))
+    if pred[0] == 1:
+        return True
+    else:
+        return False
 
 
 def respond(
@@ -17,25 +28,28 @@ def respond(
     temperature,
     top_p,
 ):
-    messages = [{"role": "municipality", "content": 지자체}]
-    messages.append({"role": "title", "content": 제목})
-    messages.append({"role": "question", "content": 질문})
-    response = ""
-
-    chat_completion = client.chat.completions.create(
-        model="tgi",
-        messages=messages,
-        stream=True,
-        max_tokens=max_tokens,
-        temperature=temperature,
-        top_p=top_p,
-    )
-
-    for message in chat_completion:
-        token = message.choices[0].delta.content
-        if token:
-            response += token
-            yield response
+    if guard_question(질문):
+        messages = [{"role": "municipality", "content": 지자체}]
+        messages.append({"role": "title", "content": 제목})
+        messages.append({"role": "question", "content": 질문})
+        response = ""
+
+        chat_completion = client.chat.completions.create(
+            model="tgi",
+            messages=messages,
+            stream=True,
+            max_tokens=max_tokens,
+            temperature=temperature,
+            top_p=top_p,
+        )
+
+        for message in chat_completion:
+            token = message.choices[0].delta.content
+            if token:
+                response += token
+                yield response
+    else:
+        yield "제가 답할 수 있는 질문이 아닌 것 같습니다. 저는 민원 게시글을 처리할 수 있어요."
 
 
 demo = gr.Interface(

question_undetector.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2dc7997100b3e24f5683560e117b15a96eced62eebb0fca245c927d3c4f01f8f
+size 1194927

requirements.txt

@@ -1,2 +1,4 @@
 huggingface_hub==0.22.2
 openai
+kiwipiepy
+scikit-learn