Update app.py

app.py

@@ -22,35 +22,33 @@ import pandas as pd
 import PyPDF2
 
 ##############################################################################
-# SERPHouse API key for web search
+# SERPHouse API key from environment variable
 ##############################################################################
-SERPHOUSE_API_KEY = "V38CNn4HXpLtynJQyOeoUensTEYoFy8PBUxKpDqAW1pawT1vfJ2BWtPQ98h6"
+SERPHOUSE_API_KEY = os.getenv("SERPHOUSE_API_KEY", "")
 
 ##############################################################################
-# [새로 추가] 사용자 메시지로부터 간단히 키워드 추출하는 함수 예시
-# - 실제 환경에 맞게 stopwords, 형태소 분석 등 고도화 가능
+# 간단한 키워드 추출 함수 (사용자 프롬프트 -> 키워드)
 ##############################################################################
 def extract_keywords(text: str, top_k: int = 5) -> str:
-    # 1) 소문자로
+    """
+    가장 간단한 예시:
+    1) 텍스트를 소문자로
+    2) 알파벳/숫자/공백 제외 문자 제거
+    3) 공백 토큰 분리
+    4) 앞 토큰 n개 추출
+    """
     text = text.lower()
-    # 2) 알파벳/숫자/공백 제외 문자 제거
     text = re.sub(r"[^a-z0-9\s]", "", text)
-    # 3) 공백단위 토큰
     tokens = text.split()
-    # 4) 우선은 앞에서 몇 개 토큰만 사용 (top_k=5)
-    #    - 필요시 stopword 제거나 빈도수 계산 후 상위 k개 추출하도록 변경 가능
     key_tokens = tokens[:top_k]
-    # 5) 공백으로 join
     return " ".join(key_tokens)
 
 ##############################################################################
-# Simple function to call the SERPHouse Live endpoint
-# https://api.serphouse.com/serp/live
+# SERPHouse Live endpoint 호출 (상위 20개의 제목을 얻음)
 ##############################################################################
 def do_web_search(query: str) -> str:
     """
-    Calls SERPHouse live endpoint with the given query (q).
-    Returns top-20 results' titles as a bullet list, or an error message.
+    SERPHouse 라이브 검색 호출, 상위 20개 결과의 'title'만 묶어서 반환.
     """
     try:
         url = "https://api.serphouse.com/serp/live"
@@ -60,11 +58,11 @@ def do_web_search(query: str) -> str:
             "lang": "en",
             "device": "desktop",
             "serp_type": "web",
-            "num_result": "20",  # [새로 추가] 상위 20개 결과
+            "num_result": "20",  # 상위 20개 결과
             "api_token": SERPHOUSE_API_KEY,
         }
         resp = requests.get(url, params=params, timeout=30)
-        resp.raise_for_status()  # Raise an exception for 4xx/5xx errors
+        resp.raise_for_status()  # 4xx/5xx 에러 시 예외
         data = resp.json()
 
         results = data.get("results", {})
@@ -72,21 +70,21 @@ def do_web_search(query: str) -> str:
         if not organic:
             return "No web search results found."
 
-        # 상위 20개 제목만 뽑아서 정리
         summary_lines = []
         for idx, item in enumerate(organic[:20], start=1):
             title = item.get("title", "No Title")
             summary_lines.append(f"{idx}. {title}")
 
-        # 20개를 \n 으로 연결
         return "\n".join(summary_lines)
     except Exception as e:
         logger.error(f"Web search failed: {e}")
         return f"Web search failed: {str(e)}"
 
 
-MAX_CONTENT_CHARS = 4000  # 너무 큰 파일을 막기 위해 최대 표시 4000자
-
+##############################################################################
+# 상수 설정
+##############################################################################
+MAX_CONTENT_CHARS = 4000  # 너무 큰 파일을 막기 위해 최대 4000자만 표시
 model_id = os.getenv("MODEL_ID", "google/gemma-3-27b-it")
 processor = AutoProcessor.from_pretrained(model_id, padding_side="left")
 model = Gemma3ForConditionalGeneration.from_pretrained(
@@ -95,16 +93,19 @@ model = Gemma3ForConditionalGeneration.from_pretrained(
     torch_dtype=torch.bfloat16,
     attn_implementation="eager"
 )
-
 MAX_NUM_IMAGES = int(os.getenv("MAX_NUM_IMAGES", "5"))
 
 
-##################################################
+##############################################################################
 # CSV, TXT, PDF 분석 함수
-##################################################
+##############################################################################
 def analyze_csv_file(path: str) -> str:
+    """
+    CSV 파일을 전체 문자열로 변환. 너무 길 경우 일부만 표시.
+    """
     try:
         df = pd.read_csv(path)
+        # 최대 50행, 10열까지만 표시
         if df.shape[0] > 50 or df.shape[1] > 10:
             df = df.iloc[:50, :10]
         df_str = df.to_string()
@@ -116,6 +117,9 @@ def analyze_csv_file(path: str) -> str:
 
 
 def analyze_txt_file(path: str) -> str:
+    """
+    TXT 파일 전문 읽기. 너무 길면 일부만 표시.
+    """
     try:
         with open(path, "r", encoding="utf-8") as f:
             text = f.read()
@@ -127,6 +131,9 @@ def analyze_txt_file(path: str) -> str:
 
 
 def pdf_to_markdown(pdf_path: str) -> str:
+    """
+    PDF → Markdown. 페이지별로 간단히 텍스트 추출.
+    """
     text_chunks = []
     try:
         with open(pdf_path, "rb") as f:
@@ -137,6 +144,7 @@ def pdf_to_markdown(pdf_path: str) -> str:
                 page_text = page.extract_text() or ""
                 page_text = page_text.strip()
                 if page_text:
+                    # 페이지별 텍스트 제한
                     if len(page_text) > MAX_CONTENT_CHARS // max_pages:
                         page_text = page_text[:MAX_CONTENT_CHARS // max_pages] + "...(truncated)"
                     text_chunks.append(f"## Page {page_num+1}\n\n{page_text}\n")
@@ -152,9 +160,9 @@ def pdf_to_markdown(pdf_path: str) -> str:
     return f"**[PDF File: {os.path.basename(pdf_path)}]**\n\n{full_text}"
 
 
-##################################################
+##############################################################################
 # 이미지/비디오 업로드 제한 검사
-##################################################
+##############################################################################
 def count_files_in_new_message(paths: list[str]) -> tuple[int, int]:
     image_count = 0
     video_count = 0
@@ -183,6 +191,13 @@ def count_files_in_history(history: list[dict]) -> tuple[int, int]:
 
 
 def validate_media_constraints(message: dict, history: list[dict]) -> bool:
+    """
+    - 비디오 1개 초과 불가
+    - 비디오와 이미지 혼합 불가
+    - 이미지 개수(MAX_NUM_IMAGES) 초과 불가
+    - <image> 태그가 있으면 태그 수와 실제 이미지 수 일치
+    - CSV, TXT, PDF 등은 여기서 제한하지 않음
+    """
     media_files = []
     for f in message["files"]:
         if re.search(r"\.(png|jpg|jpeg|gif|webp)$", f, re.IGNORECASE) or f.endswith(".mp4"):
@@ -217,14 +232,14 @@ def validate_media_constraints(message: dict, history: list[dict]) -> bool:
     return True
 
 
-##################################################
+##############################################################################
 # 비디오 처리
-##################################################
+##############################################################################
 def downsample_video(video_path: str) -> list[tuple[Image.Image, float]]:
     vidcap = cv2.VideoCapture(video_path)
     fps = vidcap.get(cv2.CAP_PROP_FPS)
     total_frames = int(vidcap.get(cv2.CAP_PROP_FRAME_COUNT))
-    frame_interval = max(int(fps), int(total_frames / 10))  
+    frame_interval = max(int(fps), int(total_frames / 10))
     frames = []
 
     for i in range(0, total_frames, frame_interval):
@@ -255,9 +270,9 @@ def process_video(video_path: str) -> list[dict]:
     return content
 
 
-##################################################
+##############################################################################
 # interleaved <image> 처리
-##################################################
+##############################################################################
 def process_interleaved_images(message: dict) -> list[dict]:
     parts = re.split(r"(<image>)", message["text"])
     content = []
@@ -277,9 +292,9 @@ def process_interleaved_images(message: dict) -> list[dict]:
     return content
 
 
-##################################################
+##############################################################################
 # PDF + CSV + TXT + 이미지/비디오
-##################################################
+##############################################################################
 def is_image_file(file_path: str) -> bool:
     return bool(re.search(r"\.(png|jpg|jpeg|gif|webp)$", file_path, re.IGNORECASE))
 
@@ -287,9 +302,12 @@ def is_video_file(file_path: str) -> bool:
     return file_path.endswith(".mp4")
 
 def is_document_file(file_path: str) -> bool:
-    return (file_path.lower().endswith(".pdf") or 
-            file_path.lower().endswith(".csv") or 
-            file_path.lower().endswith(".txt"))
+    return (
+        file_path.lower().endswith(".pdf")
+        or file_path.lower().endswith(".csv")
+        or file_path.lower().endswith(".txt")
+    )
+
 
 def process_new_user_message(message: dict) -> list[dict]:
     if not message["files"]:
@@ -321,7 +339,7 @@ def process_new_user_message(message: dict) -> list[dict]:
 
     if "<image>" in message["text"] and image_files:
         interleaved_content = process_interleaved_images({"text": message["text"], "files": image_files})
-        if content_list[0]["type"] == "text":
+        if content_list and content_list[0]["type"] == "text":
             content_list = content_list[1:]
         return interleaved_content + content_list
     else:
@@ -331,9 +349,9 @@ def process_new_user_message(message: dict) -> list[dict]:
     return content_list
 
 
-##################################################
+##############################################################################
 # history -> LLM 메시지 변환
-##################################################
+##############################################################################
 def process_history(history: list[dict]) -> list[dict]:
     messages = []
     current_user_content: list[dict] = []
@@ -360,9 +378,9 @@ def process_history(history: list[dict]) -> list[dict]:
     return messages
 
 
-##################################################
-# 메인 추론 함수
-##################################################
+##############################################################################
+# 메인 추론 함수 (web search 체크 시 자동 키워드추출->검색->결과 system msg 반영)
+##############################################################################
 @spaces.GPU(duration=120)
 def run(
     message: dict,
@@ -378,19 +396,18 @@ def run(
         return
 
     try:
-        # [새로 추가] web search 체크된 경우, 사용자가 입력한 "web_search_query" 대신
-        #             사용자의 메시지에서 키워드를 추출하여 검색
+        # web_search가 True면 => 사용자가 직접 입력한 web_search_query 대신,
+        # message["text"]를 기반으로 키워드 추출하여 검색
+        history_system_msg = None
         if use_web_search:
             user_text = message["text"]
-            # 키워드 추출
+            # 1) 키워드 추출
             ws_query = extract_keywords(user_text, top_k=5)
             logger.info(f"[Auto WebSearch Keyword] {ws_query!r}")
-            # 상위 20개 결과 가져오기
+            # 2) 상위 20개 결과 불러오기
             ws_result = do_web_search(ws_query)
-            # 검색된 20개 제목을 system 메시지에 추가
+            # 3) 이를 system 메시지로 추가
             system_search_content = f"[Search top-20 Titles Based on user prompt]\n{ws_result}\n"
-            # system 메시지로 추가
-            # (LLM이 이 정보를 참고하도록)
             if system_search_content.strip():
                 history_system_msg = {
                     "role": "system",
@@ -401,25 +418,26 @@ def run(
                     "role": "system",
                     "content": [{"type": "text", "text": "No web search results"}]
                 }
-        else:
-            history_system_msg = None
 
+        # 기존 system prompt
         messages = []
         if system_prompt:
             messages.append({"role": "system", "content": [{"type": "text", "text": system_prompt}]})
-        # 만약 web search가 있었다면, 그 결과를 추가 system 메시지로 삽입
+        # web 검색 결과 system msg
         if history_system_msg:
             messages.append(history_system_msg)
 
+        # 이전 대화이력(assistant/user)
         messages.extend(process_history(history))
         
+        # 새 유저 메시지 변환
         user_content = process_new_user_message(message)
         for item in user_content:
             if item["type"] == "text" and len(item["text"]) > MAX_CONTENT_CHARS:
                 item["text"] = item["text"][:MAX_CONTENT_CHARS] + "\n...(truncated)..."
-        
         messages.append({"role": "user", "content": user_content})
 
+        # LLM 입력 생성
         inputs = processor.apply_chat_template(
             messages,
             add_generation_prompt=True,
@@ -442,15 +460,16 @@ def run(
         for new_text in streamer:
             output += new_text
             yield output
-            
+
     except Exception as e:
         logger.error(f"Error in run: {str(e)}")
         yield f"죄송합니다. 오류가 발생했습니다: {str(e)}"
 
 
-
+##############################################################################
+# 예시들 (한글화)
+##############################################################################
 examples = [
-
     [
         {
             "text": "두 PDF 파일 내용을 비교하라.",
@@ -458,7 +477,7 @@ examples = [
             "files": [
                 "assets/additional-examples/before.pdf",
                 "assets/additional-examples/after.pdf",
-            ],            
+            ],
         }
     ],
     [
@@ -466,37 +485,37 @@ examples = [
             "text": "CSV 파일 내용을 요약, 분석하라",
             "files": ["assets/additional-examples/sample-csv.csv"],
         }
-    ],    
+    ],
     [
         {
             "text": "이 영상의 내용을 설명하라",
             "files": ["assets/additional-examples/tmp.mp4"],
         }
-    ],    
+    ],
     [
         {
             "text": "표지 내용을 설명하고 글자를 읽어주세요.",
             "files": ["assets/additional-examples/maz.jpg"],
         }
-    ],    
+    ],
     [
         {
             "text": "이미 이 영양제를 <image> 가지고 있고, 이 제품 <image>을 새로 사려 합니다. 함께 섭취할 때 주의해야 할 점이 있을까요?",
             "files": ["assets/additional-examples/pill1.png", "assets/additional-examples/pill2.png"],
         }
-    ],    
+    ],
     [
         {
             "text": "이 적분��� 풀어주세요.",
             "files": ["assets/additional-examples/4.png"],
         }
-    ],    
+    ],
     [
         {
             "text": "이 티켓은 언제 발급된 것이고, 가격은 얼마인가요?",
             "files": ["assets/additional-examples/2.png"],
         }
-    ],    
+    ],
     [
         {
             "text": "이미지들의 순서를 바탕으로 짧은 이야기를 만들어 주세요.",
@@ -520,24 +539,19 @@ examples = [
             "text": "동일한 막대 그래프를 그리는 matplotlib 코드를 작성해주세요.",
             "files": ["assets/additional-examples/barchart.png"],
         }
-    ],    
-
+    ],
     [
         {
             "text": "이 세계에서 살고 있을 생물들을 상상해서 묘사해주세요.",
             "files": ["assets/sample-images/08.png"],
         }
     ],
-
-
     [
         {
             "text": "이미지에 있는 텍스트를 그대로 읽어서 마크다운 형태로 적어주세요.",
             "files": ["assets/additional-examples/3.png"],
         }
     ],
-
-
     [
         {
             "text": "이 표지판에는 무슨 문구가 적혀 있나요?",
@@ -550,10 +564,12 @@ examples = [
             "files": ["assets/sample-images/03.png"],
         }
     ],
-
 ]
 
 
+##############################################################################
+# Gradio UI (Blocks) 구성
+##############################################################################
 css = """
 body {
     background: linear-gradient(135deg, #667eea, #764ba2);
@@ -626,9 +642,9 @@ with gr.Blocks(css=css, title="Vidraft-Gemma-3-27B") as demo:
                 web_search_checkbox = gr.Checkbox(
                     label="Web Search",
                     value=False,
-                    info="Check to enable a SERPHouse web search before the chat reply"
+                    info="Check to enable a SERPHouse web search (auto keywords) before the chat reply"
                 )
-            # [중요] web_search_text는 사실상 사용 안 함 (자동추출로 검색)
+            # 실제로는 자동추출. 아래 textbox는 미사용.
             web_search_text = gr.Textbox(
                 lines=1,
                 label="(Unused) Web Search Query",
@@ -653,8 +669,9 @@ with gr.Blocks(css=css, title="Vidraft-Gemma-3-27B") as demo:
                 value=2000,
             )
 
-            gr.Markdown("<br><br>")
+            gr.Markdown("<br><br>")  # spacing
 
+        # Main ChatInterface to the right
         with gr.Column(scale=7):
             chat = gr.ChatInterface(
                 fn=run,
@@ -673,7 +690,7 @@ with gr.Blocks(css=css, title="Vidraft-Gemma-3-27B") as demo:
                     system_prompt_box,
                     max_tokens_slider,
                     web_search_checkbox,
-                    web_search_text,  # 실제로는 사용 안함
+                    web_search_text,  # 실제로는 auto search
                 ],
                 stop_btn=False,
                 title="Vidraft-Gemma-3-27B",
@@ -689,7 +706,7 @@ with gr.Blocks(css=css, title="Vidraft-Gemma-3-27B") as demo:
             gr.Markdown("### Example Inputs (click to load)")
             gr.Examples(
                 examples=examples,
-                inputs=[],  # Gradio가 dataset에 연결할 inputs가 없으므로 빈 리스트
+                inputs=[],  # 링크할 inputs가 없으므로 빈 리스트
                 cache_examples=False
             )