Add Colab notebook converted to app.py

app.py

@@ -1,70 +1,134 @@
-import gradio as gr
-from huggingface_hub import InferenceClient
-
-
-def respond(
-    message,
-    history: list[dict[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-    hf_token: gr.OAuthToken,
-):
+!pip install -q -U transformers peft accelerate bitsandbytes
+
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+from peft import PeftModel
+from google.colab import drive
+
+# ============================================
+# 1️⃣ 드라이브 마운트
+# ============================================
+drive.mount('/content/drive')
+
+# ============================================
+# 2️⃣ 환경 설정
+# ============================================
+BASE_MODEL = "beomi/Llama-3-Open-Ko-8B"
+LORA_PATH = "/content/drive/MyDrive/at_last"
+
+print("🚀 모델 로드 중...")
+model = AutoModelForCausalLM.from_pretrained(
+    BASE_MODEL,
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+    trust_remote_code=True
+)
+
+tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
+tokenizer.pad_token = tokenizer.eos_token
+
+print("🔗 LoRA 병합 중...")
+model = PeftModel.from_pretrained(model, LORA_PATH, is_local=True)
+
+# ✅ <|eot_id|> 토큰을 EOS로 지정
+model.config.eos_token_id = tokenizer.eos_token_id
+model.config.pad_token_id = tokenizer.pad_token_id
+
+print("✅ 모델 + LoRA 준비 완료!")
+
+from transformers import StoppingCriteria, StoppingCriteriaList
+
+class StopOnTokens(StoppingCriteria):
+    def __init__(self, stop_ids):
+        self.stop_ids = stop_ids
+
+    def __call__(self, input_ids, scores, **kwargs):
+        last_token = input_ids[0, -1].item()
+        return last_token in self.stop_ids
+
+
+# ✅ 종료 토큰 후보를 모두 등록
+stop_words = ["<|eot|>", "</s>", "<|end_of_text|>"]
+stop_ids = [tokenizer.convert_tokens_to_ids(w) for w in stop_words if tokenizer.convert_tokens_to_ids(w) is not None]
+stopping_criteria = StoppingCriteriaList([StopOnTokens(stop_ids)])
+
+stopping_criteria = StoppingCriteriaList([StopOnTokens(stop_ids)])
+
+# ============================================
+# 3️⃣ 프롬프트 빌드 함수
+# ============================================
+AI_PERSONALITY = """
+너는 사용자의 말을 진심으로 들어주는 친구야.
+사용자가 대화를 걸면 자연스럽고 일상적인 톤으로 대답해.
+장황하지 말고, 공감하면서 짧고 따뜻하게 말할 것.
+너는 사용자의 요청을 정확히 이해하고, 현실적인 답변을 제공하는 친근한 친구야.
+농담과 공감을 섞되, 요청을 회피하지 않고 명확히 답변해야 해.
+"""
+
+def build_prompt_full_history(history):
     """
-    For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
+    - history는 user/assistant 모든 대화 포함
+    - 마지막 user 발화만 generate 대상
     """
-    client = InferenceClient(token=hf_token.token, model="openai/gpt-oss-20b")
+    prompt = "<|begin_of_text|>\n" + AI_PERSONALITY.strip() + "\n\n"
+    for turn in history:
+        role = turn["role"]
+        content = turn["content"].strip()
+        prompt += f"<|start_header_id|>{role}<|end_header_id|>\n{content}<|eot|>\n"
 
-    messages = [{"role": "system", "content": system_message}]
+    # 마지막 user 이후에 assistant placeholder 추가
+    prompt += "<|start_header_id|>assistant<|end_header_id|>\n"
+    return prompt
 
-    messages.extend(history)
 
-    messages.append({"role": "user", "content": message})
 
-    response = ""
+# ============================================
+# 4️⃣ 대화 루프
+# ============================================
+history = []
+add_header = True  # 첫 턴만 personality 포함
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        choices = message.choices
-        token = ""
-        if len(choices) and choices[0].delta.content:
-            token = choices[0].delta.content
+while True:
+    user_input = input("👤 사용자: ").strip()
+    if user_input.lower() in ["종료", "exit", "quit"]:
+        print("🛑 대화 종료!")
+        break
 
-        response += token
-        yield response
+    history.append({"role": "user", "content": user_input})
+    prompt = build_prompt_full_history(history)
+    add_header = False  # 이후에는 personality 중복 방지
 
+    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-chatbot = gr.ChatInterface(
-    respond,
-    type="messages",
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
+    with torch.no_grad():
+      output = model.generate(
+          **inputs,
+          max_new_tokens=256,
+          temperature=0.6,
+          top_p=0.9,
+          repetition_penalty=1.1,
+          pad_token_id=tokenizer.eos_token_id,
+          eos_token_id=tokenizer.eos_token_id,
+          stopping_criteria=stopping_criteria
+      )
+
+    response_full = tokenizer.decode(
+        output[0][inputs["input_ids"].shape[1]:],
+        skip_special_tokens=True
+    )
+
+    response = response_full.split("<|eot|>")[0].strip()
+
+
+    # <|eot_id|> 기준으로 자르기
+    if "<|eot_id|>" in response_full:
+        response = response_full.split("<|eot_id|>")[0].strip()
+    else:
+        response = response_full.strip()
 
-with gr.Blocks() as demo:
-    with gr.Sidebar():
-        gr.LoginButton()
-    chatbot.render()
+    print(f"🤖 AI: {response}\n")
 
+    history.append({"role": "assistant", "content": response})
+    if len(history) > 10:
+        history = history[-10:]
 
-if __name__ == "__main__":
-    demo.launch()