Update api.py

api.py

@@ -6,10 +6,10 @@ import asyncio
 from fastapi import FastAPI, Request
 from fastapi.responses import StreamingResponse
 import sentencepiece as spm
-from typing import List
 import re
-app = FastAPI()
-dialogue_history = []
+app = FastAPI()    
+
+
 from fastapi.middleware.cors import CORSMiddleware
 
 origins = [
@@ -199,14 +199,10 @@ def generate_text_greedy_strong(model, prompt, max_len=100, max_gen=98,
         next_token = np.argmax(next_logits)
         generated.append(int(next_token))  
 
-        decoded = sp.decode(generated)
-
-# <start>와 <sep> 제거
-        for t in ["<start>", "<sep>", "<end>"]:
-            decoded = decoded.replace(t, "")
-
-# 입력 프롬프트도 제거
-        decoded = decoded.replace(prompt, "").strip()
+        decoded = sp.decode(generated)  
+        for t in ["<start>", "<sep>", "<end>"]:  
+            decoded = decoded.replace(t, "")  
+        decoded = decoded.strip()
 
         if len(generated) >= min_len and (next_token == end_id or decoded.endswith(('요', '다', '.', '!', '?'))):  
             if is_greedy_response_acceptable(decoded):
@@ -289,14 +285,6 @@ def summarize_from_wikipedia(query, top_n=3):
     raw_summary = get_wikipedia_summary(query)
     return textrank_summarize(raw_summary, top_n=top_n)
 
-def build_contexted_prompt(history: List[str], user_input: str):
-    # 최근 대화 3개를 합쳐서 요약
-    recent = ' '.join(history[-3:])
-    summary = textrank_summarize(recent, top_n=2)
-    # 요약 + 최신 사용자 입력을 같이 던져줌
-    prompt = f"{summary} {user_input}"
-    return prompt
-
 # 의도 분류기
 def simple_intent_classifier(text):
     text = text.lower()
@@ -321,10 +309,8 @@ def parse_math_question(text):
     except:
         return "계산할 수 없는 수식이에요. 다시 한번 확인해 주세요!"
 
-# respond 함수 수정
+# 최종 응답 함수
 def respond(input_text):
-    global dialogue_history
-
     intent = simple_intent_classifier(input_text)
 
     if "이름" in input_text:
@@ -334,38 +320,22 @@ def respond(input_text):
         return "저는 Ector.V라고 해요."
 
     if intent == "수학질문":
-        dialogue_history.append(f"사용자: {input_text}")
-        response = parse_math_question(input_text)
-        dialogue_history.append(f"Ector: {response}")
-        return response
+        return parse_math_question(input_text)
 
     if intent == "인사":
-        response = "반가워요! 무엇을 도와드릴까요?"
-        dialogue_history.append(f"사용자: {input_text}")
-        dialogue_history.append(f"Ector: {response}")
-        return response
+        return "반가워요! 무엇을 도와드릴까요?"
 
     if intent == "정보질문":
         keyword = re.sub(r"(에 대해|에 대한|에 대해서)?\s*(설명해줘|알려줘|뭐야|개념|정의|정보)?", "", input_text).strip()
         if not keyword:
             return "어떤 주제에 대해 궁금한가요?"
         summary = summarize_from_wikipedia(keyword)
-        response = f"{summary}\n다른 궁금한 점 있으신가요?"
-        dialogue_history.append(f"사용자: {input_text}")
-        dialogue_history.append(f"Ector: {response}")
-        return response
+        return f"{summary}\n다른 궁금한 점 있으신가요?"
 
-    # 일상 대화: 요약 기반 컨텍스트 생성
-    contexted_prompt = build_contexted_prompt(dialogue_history, input_text)
-    response = generate_text_greedy_strong(model, contexted_prompt)
-
-    # fallback
+    # 일상 대화: 샘플링 + fallback
+    response = generate_text_greedy_strong(model, input_text)
     if not is_valid_response(response) or mismatch_tone(input_text, response):
-        response = generate_text_greedy_strong(model, contexted_prompt)
-
-    # 히스토리 추가
-    dialogue_history.append(f"사용자: {input_text}")
-    dialogue_history.append(f"Ector: {response}")
+        response = generate_text_greedy_strong(model, input_text)
     return response
 
 @app.get("/generate", response_class=PlainTextResponse)