add "start with localhost", code refactored

.vscode/settings.json

@@ -0,0 +1,6 @@
+{
+  "python.analysis.extraPaths": [
+    "."
+  ],
+  "python.envFile": "${workspaceFolder}/.env"
+}

app.py

@@ -1,145 +1,21 @@
-from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
 import gradio as gr
-import torch
-import re
-import os
 
-print("[torch] is available:", torch.cuda.is_available())
-print("[device] default:", torch.device("cuda" if torch.cuda.is_available() else "cpu"))
+from core.make_pipeline import MakePipeline
+from core.context_manager import ContextManager
+from core.launch_gradio import create_interface
 
-# 모델 로드
-model_id = "naver-hyperclovax/HyperCLOVAX-SEED-Text-Instruct-1.5B"
+###########################
+# Upload to Huggling Face #
+###########################
 
-# 허깅 페이스 secret에 등록된 토큰 로드
-access_token = os.environ.get("HF_TOKEN")
+# 모델 불러오기
+makePipeline = MakePipeline()
+makePipeline.build("hf")
 
-tokenizer = AutoTokenizer.from_pretrained(model_id, token=access_token)
-model = AutoModelForCausalLM.from_pretrained(
-    model_id,
-    token=access_token
-)
-model.eval()
-llm = pipeline(
-    "text-generation",
-    model=model,
-    tokenizer=tokenizer,
-    torch_dtype=torch.float16
-)
+# 채팅 기록 관리자
+ctx = ContextManager()
 
-# 챗봇 프롬프트 생성
-def build_prompt(history, user_msg, user_name="User", bot_name="Tanjiro"):
-    with open("assets/prompt/init.txt", "r", encoding="utf-8") as f:
-        prompt = f.read().strip()
-
-    for turn in history[-16:]:
-        role = user_name if turn["role"] == "user" else bot_name
-        prompt += f"{role}: {turn['text']}\n"
-
-    prompt += f"{user_name}: {user_msg}\n"
-    prompt += f"{bot_name}:"
-    return prompt
-
-# 출력에서 응답 추출
-def extract_response(full_text, prompt, bot_name="Tanjiro"):
-    if full_text.startswith(prompt):
-        reply = full_text[len(prompt):].strip()
-    else:
-        reply = full_text.split(f"{bot_name}:")[-1].strip()
-    user_token = "\nUser:"
-    if user_token in reply:
-        reply = reply.split(user_token)[0].strip()
-    return reply
-
-# 출력 생성 함수
-def character_chat(user_msg, history):
-    print("[debug] generationg...")
-    prompt = build_prompt(history, user_msg)
-    outputs = llm(
-        prompt,
-        do_sample=True,
-        max_new_tokens=96,
-        temperature=0.7,
-        top_p=0.9,
-        repetition_penalty=1.05,
-        eos_token_id=tokenizer.eos_token_id,
-        return_full_text=True
-    )
-    full_text = outputs[0]['generated_text']
-    response = extract_response(full_text, prompt)
-    return response
-
-# 중단된 응답 여부 검사
-def is_truncated_response(text: str) -> bool:
-    return re.search(r"[.?!…\u2026\u2639\u263A\u2764\uD83D\uDE0A\uD83D\uDE22]$", text.strip()) is None
-
-# 답변 유효성 검사
-def is_valid_response(text: str, bot_name="Tanjiro", user_name="User") -> bool:
-    if user_name + ":" in text:
-        return False
-    if bot_name + ":" in text:
-        return False
-    return True
-
-# 답변 형식 정리
-def clean_response(text: str, bot_name="Tanjiro"):
-    return re.sub(rf"{bot_name}:\\s*", "", text).strip()
-
-# Gradio 인터페이스
-with gr.Blocks(css="""
-.chat-box { max-height: 500px; overflow-y: auto; padding: 10px; border: 1px solid #ccc; border-radius: 10px; }
-.bubble-left { background-color: #f1f0f0; border-radius: 10px; padding: 10px; margin: 5px; max-width: 70%; float: left; clear: both; }
-.bubble-right { background-color: #d1e7ff; border-radius: 10px; padding: 10px; margin: 5px; max-width: 70%; float: right; clear: both; text-align: right; }
-.reset-btn-container { text-align: right; margin-bottom: 10px; }
-""") as demo:
-    gr.Markdown("### 탄지로와 대화하기")
-    with gr.Column():
-        with gr.Row():
-            gr.Markdown("")
-            reset_btn = gr.Button("🔁 대화 초기화", elem_classes="reset-btn-container", scale=1)
-        chat_output = gr.HTML(elem_id="chat-box")
-        user_input = gr.Textbox(label="메시지 입력", placeholder="탄지로에게 말을 걸어보세요")
-        state = gr.State([])
-
-    def render_chat(history):
-        html = ""
-        for item in history:
-            if item["role"] == "user":
-                html += f"<div class='bubble-right'>{item['text']}</div>"
-            elif item["role"] == "bot":
-                html += f"<div class='bubble-left'>{item['text']}</div>"
-        return gr.update(value=html)
-
-    def on_submit(user_msg, history):
-        history.append({"role": "user", "text": user_msg})
-        html = render_chat(history)
-        yield html, "", history
-
-        #응답 생성
-        while True:
-            response = character_chat(user_msg, history)
-            if is_valid_response(response):
-                break
-        response = clean_response(response)
-        history.append({"role": "bot", "text": response})
-
-        #중간에 응답이 끊긴 경우 추가 생성
-        if is_truncated_response(response):
-            while True:
-                continuation = character_chat(response, history)
-                if is_valid_response(continuation):
-                    break
-            continuation = clean_response(continuation)
-            history.append({"role": "bot", "text": continuation})
-
-        html = render_chat(history)
-        yield html, "", history
-
-    def reset_chat():
-        return gr.update(value=""), "", []
-
-    user_input.submit(on_submit, inputs=[user_input, state], outputs=[chat_output, user_input, state], queue=True)
-    reset_btn.click(reset_chat, inputs=None, outputs=[chat_output, user_input, state])
-    
-    #허깅페이스에서 실행
-    demo.launch()
+# Gradio 인터페이스 시작
+demo = create_interface(ctx, makePipeline)
+demo.launch()
     

app/main.py

@@ -0,0 +1,19 @@
+from core.make_pipeline import MakePipeline
+from core.context_manager import ContextManager
+from core.launch_gradio import create_interface
+
+########################
+# Start with localhost #
+########################
+
+if __name__ == "__main__":
+    # 모델 불러오기
+    makePipeline = MakePipeline()
+    makePipeline.build("hf")
+
+    # 채팅 기록 관리자
+    ctx = ContextManager()
+
+    # Gradio 인터페이스 시작
+    demo = create_interface(ctx, makePipeline)
+    demo.launch()

core/context_manager.py

@@ -0,0 +1,29 @@
+class ContextManager:
+    # 전역 상수 설정
+    USER_NAME = "User"
+    BOT_NAME = "Tanjiro"
+
+    def __init__(self):
+        self.user_name = self.USER_NAME
+        self.bot_name = self.BOT_NAME
+        self.history = []
+
+    def getUserName(self) -> str:
+        return self.user_name
+    
+    def getBotName(self) -> str:
+        return self.bot_name
+    
+    def getHistory(self) -> str:
+        return self.history
+    
+    def setHistory(self, new_history: list):
+        self.history = new_history
+    
+    # 대화 기록을 history에 추가
+    def addHistory(self, role: str, text: str):
+        self.history.append({"role": role, "text": text})
+
+    # 대화 기록 초기화
+    def clearHistory(self):
+        self.history = []

core/launch_gradio.py

@@ -0,0 +1,56 @@
+import gradio as gr
+from context_manager import ContextManager
+from make_pipeline import MakePipeline
+from make_reply import generate_reply
+
+def create_interface(ctx: ContextManager, makePipeline: MakePipeline):
+    with gr.Blocks(css="""
+    .chat-box { max-height: 500px; overflow-y: auto; padding: 10px; border: 1px solid #ccc; border-radius: 10px; }
+    .bubble-left { background-color: #f1f0f0; border-radius: 10px; padding: 10px; margin: 5px; max-width: 70%; float: left; clear: both; }
+    .bubble-right { background-color: #d1e7ff; border-radius: 10px; padding: 10px; margin: 5px; max-width: 70%; float: right; clear: both; text-align: right; }
+    .reset-btn-container { text-align: right; margin-bottom: 10px; }
+    """) as demo:
+        gr.Markdown("### 탄지로와 대화하기")
+
+        with gr.Column():
+            with gr.Row():
+                gr.Markdown("")
+                reset_btn = gr.Button("🔁 대화 초기화", elem_classes="reset-btn-container", scale=1)
+            chat_output = gr.HTML(elem_id="chat-box")
+            user_input = gr.Textbox(label="메시지 입력", placeholder="탄지로에게 말을 걸어보세요")
+            state = gr.State(ctx)
+
+        # history 읽어서 화면에 뿌리는 역할
+        def render_chat(ctx: ContextManager):
+            html = ""
+            for item in ctx.getHistory():
+                if item["role"] == ctx.getUserName():
+                    html += f"<div class='bubble-right'>{item['text']}</div>"
+                elif item["role"] == ctx.getBotName():
+                    html += f"<div class='bubble-left'>{item['text']}</div>"
+            return gr.update(value=html)
+
+        def on_submit(user_msg: str, ctx: ContextManager):
+            # 사용자 메세지 추가
+            ctx.addMessage("user", user_msg)
+
+            # 유저 답변을 포함한 HTML 렌더링
+            html = render_chat(ctx)
+            yield html, "", ctx.getHistory()
+
+            # 봇 응답 생성
+            generate_reply(ctx, makePipeline, user_msg)
+
+            # 봇 답변을 포함한 HTML 렌더링
+            html = render_chat(ctx)
+            yield html, "", ctx.getHistory()
+
+        # history 초기화
+        def reset_chat():
+            ctx.resetHistory()
+            return gr.update(value=""), "", ctx.getHistory()
+
+        user_input.submit(on_submit, inputs=[user_input, state], outputs=[chat_output, user_input, state], queue=True)
+        reset_btn.click(reset_chat, inputs=None, outputs=[chat_output, user_input, state])
+
+        return demo

core/make_pipeline.py

@@ -0,0 +1,70 @@
+from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
+import torch
+import os
+
+class MakePipeline:
+    # 모델명
+    MODEL_ID = "naver-hyperclovax/HyperCLOVAX-SEED-Text-Instruct-1.5B"
+    
+    # 변수초기화
+    # model_id
+    # tokenizer
+    # llm
+    def __init__(self, model_id: str = MODEL_ID):
+        print("[torch] is available:", torch.cuda.is_available())
+        print("[device] default:", torch.device("cuda" if torch.cuda.is_available() else "cpu"))
+        self.model_id = model_id
+        self.tokenizer = None
+        self.llm = None
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+    # 모델 불러오기
+    def build(self, type: str):
+        if(type == 'hf'):
+            # 허깅 페이스 secret에 등록된 토큰 로드
+            access_token = os.environ.get("HF_TOKEN")
+        else:
+            # 로컬 실행시 token.txt에서 토큰 로드
+            with open("token.txt", "r") as f:
+                access_token = f.read().strip()
+
+        tokenizer = AutoTokenizer.from_pretrained(self.model_id, token=access_token)
+        model = AutoModelForCausalLM.from_pretrained(self.model_id, token=access_token)
+        self.tokenizer = tokenizer
+
+        # 허깅 페이스 업로드 시 f16 사용 안 함
+        if(type == 'hf'):
+            llm = pipeline(
+                "text-generation",
+                model=model,
+                tokenizer=tokenizer,
+            )
+
+        else:
+            model.eval()
+            llm = pipeline(
+                "text-generation",
+                model=model,
+                tokenizer=tokenizer,
+                torch_dtype=torch.float16
+            )
+            if torch.cuda.is_available():
+                model.to("cuda")
+
+        self.llm = llm
+    
+    # 모델 출력 생성 함수
+    def character_chat(self, prompt):
+        print("[debug] generating...")
+        outputs = self.llm(
+            prompt,
+            do_sample=True,
+            max_new_tokens=96,
+            temperature=0.7,
+            top_p=0.9,
+            repetition_penalty=1.05,
+            eos_token_id=self.tokenizer.eos_token_id,
+            return_full_text=True
+        )
+        full_text = outputs[0]['generated_text']
+        return full_text

core/make_reply.py

@@ -0,0 +1,68 @@
+import re
+
+from core.make_pipeline import MakePipeline
+
+# 생성된 모든 봇 응답 기록
+def generate_reply(ctx, makePipeLine: MakePipeline, user_msg):
+
+    # 최초 응답
+    response = generate_valid_response(ctx, makePipeLine, user_msg)
+    ctx.addHistory("bot", response)
+
+    # 응답이 끊겼다면 추가 생성
+    if is_truncated_response(response):
+        continuation = generate_valid_response(ctx, makePipeLine, response)
+        ctx.addHistory("bot", continuation)
+
+# 봇 응답 1회 생성
+def generate_valid_response(ctx, makePipeline: MakePipeline, user_msg: str) -> str:
+    user_name = ctx.getUserName()
+    bot_name = ctx.getBotName()
+
+    while True:
+        prompt = build_prompt(ctx.getHistory(), user_msg, user_name, bot_name)
+        full_text = makePipeline.character_chat(prompt)
+        response = extract_response(full_text, prompt, user_name, bot_name)
+        if is_valid_response(response, user_name, bot_name):
+            break
+    return clean_response(response, bot_name)
+
+# 입력 프롬프트 정리
+def build_prompt(history, user_msg, user_name, bot_name):
+    with open("assets/prompt/init.txt", "r", encoding="utf-8") as f:
+        prompt = f.read().strip()
+
+    for turn in history[-16:]:
+        role = user_name if turn["role"] == "user" else bot_name
+        prompt += f"{role}: {turn['text']}\n"
+
+    prompt += f"{user_name}: {user_msg}\n"
+    prompt += f"{bot_name}:"
+    return prompt
+
+# 출력에서 응답 추출
+def extract_response(full_text, prompt, user_name, bot_name):
+    if full_text.startswith(prompt):
+        reply = full_text[len(prompt):].strip()
+    else:
+        reply = full_text.split(f"{bot_name}:")[-1].strip()
+    user_token = f"\n{user_name}:"
+    if user_token in reply:
+        reply = reply.split(user_token)[0].strip()
+    return reply
+
+# 응답 유효성 검사
+def is_valid_response(text: str, user_name, bot_name) -> bool:
+    if user_name + ":" in text:
+        return False
+    if bot_name + ":" in text:
+        return False
+    return True
+
+# 응답 형식 정리
+def clean_response(text: str, bot_name):
+    return re.sub(rf"{bot_name}:\\s*", "", text).strip()
+
+# 중단된 응답 여부 검사
+def is_truncated_response(text: str) -> bool:
+    return re.search(r"[.?!…\u2026\u2639\u263A\u2764\uD83D\uDE0A\uD83D\uDE22]$", text.strip()) is None

howToStart.txt

@@ -7,11 +7,14 @@ python -m venv venv
 2. activate venv
 venv\Scripts\activate
 
-3. install requirements
-pip install -r requirements.txt
+3. install requirements_local
+pip install -r requirements_local.txt
+!! not `requirements` !!
 
 3.2. recieve hugging face token
 and write it into `./token.txt`
 
 4. start server
-python app\main.py
+python app\main.py
+
+5. connect to localhost

requirements.txt

@@ -25,8 +25,8 @@ markdown-it-py==3.0.0
 MarkupSafe==3.0.2
 mdurl==0.1.2
 mpmath==1.3.0
-networkx==3.5
-numpy==2.3.1
+networkx==3.4.2
+numpy==1.26.4
 orjson==3.10.18
 packaging==25.0
 pandas==2.3.1
@@ -57,7 +57,7 @@ sympy==1.13.3
 tiktoken==0.9.0
 tokenizers==0.21.2
 tomlkit==0.13.3
-torch==2.7.1+cu118
+torch>=2.2,<2.8
 tqdm==4.67.1
 transformers==4.53.2
 typer==0.16.0
@@ -66,4 +66,4 @@ typing_extensions==4.14.1
 tzdata==2025.2
 urllib3==2.5.0
 uvicorn==0.35.0
-websockets==15.0.1
+websockets==15.0.1

requirements_local.txt

@@ -0,0 +1,69 @@
+aiofiles==24.1.0
+annotated-types==0.7.0
+anyio==4.9.0
+blobfile==3.0.0
+Brotli==1.1.0
+certifi==2025.7.14
+charset-normalizer==3.4.2
+click==8.2.1
+colorama==0.4.6
+fastapi==0.116.1
+ffmpy==0.6.0
+filelock==3.18.0
+fsspec==2025.5.1
+gradio==5.37.0
+gradio_client==1.10.4
+groovy==0.1.2
+h11==0.16.0
+httpcore==1.0.9
+httpx==0.28.1
+huggingface-hub==0.33.4
+idna==3.10
+Jinja2==3.1.6
+lxml==6.0.0
+markdown-it-py==3.0.0
+MarkupSafe==3.0.2
+mdurl==0.1.2
+mpmath==1.3.0
+networkx==3.5
+numpy==2.3.1
+orjson==3.10.18
+packaging==25.0
+pandas==2.3.1
+pillow==11.3.0
+protobuf==6.31.1
+pycryptodomex==3.23.0
+pydantic==2.11.7
+pydantic_core==2.33.2
+pydub==0.25.1
+Pygments==2.19.2
+python-dateutil==2.9.0.post0
+python-multipart==0.0.20
+pytz==2025.2
+PyYAML==6.0.2
+regex==2024.11.6
+requests==2.32.4
+rich==14.0.0
+ruff==0.12.3
+safehttpx==0.1.6
+safetensors==0.5.3
+semantic-version==2.10.0
+sentencepiece==0.2.0
+shellingham==1.5.4
+six==1.17.0
+sniffio==1.3.1
+starlette==0.47.1
+sympy==1.13.3
+tiktoken==0.9.0
+tokenizers==0.21.2
+tomlkit==0.13.3
+torch==2.7.1+cu118
+tqdm==4.67.1
+transformers==4.53.2
+typer==0.16.0
+typing-inspection==0.4.1
+typing_extensions==4.14.1
+tzdata==2025.2
+urllib3==2.5.0
+uvicorn==0.35.0
+websockets==15.0.1

todo.txt

@@ -2,4 +2,5 @@
 누적되는 대화를 프롬프트로 다시 사용할 때 적절한 길이 찾기
 이미지 출력 기능 추가
 출력 프롬프트 파싱 기능 추가
-입력 프롬프트 파싱 기능
+입력 프롬프트 파싱 기능
+character_chat 함수 설정 json으로 만들기, gradio에서 설정 가능하게 만들기