Upload 2 files

api_server (1).py

@@ -0,0 +1,124 @@
+# api_server.py
+
+from fastapi import FastAPI, HTTPException
+from fastapi.middleware.cors import CORSMiddleware
+from pydantic import BaseModel
+from transformers import GPT2LMHeadModel, GPT2Tokenizer
+import torch
+
+# =================================================================
+# 1. 應用程式初始化與模型載入
+# =================================================================
+app = FastAPI(title="GPT-2 Nursing Completion API")
+
+# 設置 CORS：允許前端頁面 (localhost 或您的服務器 IP) 訪問
+# ⚠️ 注意：在生產環境中，請將 "http://localhost:5500" 替換為您的前端域名！
+origins = [
+    #"http://localhost:5500",  # 假設您使用 VS Code Live Server 或類似工具
+    #"http://127.0.0.1:5500",
+    "https://marcoleung052.github.io/NursingRecordCompletion_train//step7/%E8%AD%B7%E7%90%86%E7%B4%80%E9%8C%84%E7%B3%BB%E7%B5%B1demo.html",
+    "*" # 為了測試方便，暫時允許所有來源
+]
+
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=origins,
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+# 全局變數用於存儲模型和分詞器
+tokenizer = None
+model = None
+MODEL_PATH = "gpt2" # 這裡可以替換為您微調後的模型資料夾路徑
+
+@app.on_event("startup")
+async def load_model():
+    """在應用啟動時載入 GPT-2 模型"""
+    global tokenizer, model
+    try:
+        # 載入分詞器
+        tokenizer = GPT2Tokenizer.from_pretrained(MODEL_PATH) 
+        
+        # 載入預訓練模型或您微調的模型權重
+        # 如果您的記憶體允許，可以考慮使用 GPU
+        # device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        model = GPT2LMHeadModel.from_pretrained(MODEL_PATH)
+        # model.to(device)
+        model.eval() # 設定為評估模式
+        
+        print(f"✅ GPT-2 模型 {MODEL_PATH} 載入成功！")
+    except Exception as e:
+        print(f"❌ 模型載入失敗，請檢查 MODEL_PATH 或依賴庫是否安裝：{e}")
+
+
+# =================================================================
+# 2. API 請求與響應格式
+# =================================================================
+class PredictionRequest(BaseModel):
+    """前端發送的請求體格式"""
+    prompt: str
+    patient_id: str | None = None
+    model: str | None = "gpt2-nursing"
+
+class PredictionResponse(BaseModel):
+    """後端回傳的響應體格式"""
+    completions: list[str]
+
+# =================================================================
+# 3. 核心 API 端點 (已修改為生成 3 個序列)
+# =================================================================
+@app.post("/api/predict", response_model=PredictionResponse)
+def predict_completion(request: PredictionRequest):
+    """根據輸入提示詞生成 DART 護理紀錄"""
+    if model is None or tokenizer is None:
+        raise HTTPException(status_code=503, detail="AI 模型服務尚未準備就緒，請檢查後端日誌。")
+    
+    input_text = request.prompt
+    if len(input_text) > 512:
+        raise HTTPException(status_code=400, detail="輸入過長，請限制在 512 個字元內。")
+
+    try:
+        input_ids = tokenizer.encode(input_text, return_tensors='pt', truncation=True)
+        
+        # 🔥 核心修改：設置 num_return_sequences=3 來生成多個候選結果
+        output = model.generate(
+            input_ids, 
+            max_length=len(input_text) + 150,
+            num_return_sequences=3,            # <--- 輸出 3 個不同的補全結果
+            no_repeat_ngram_size=3,
+            do_sample=True,
+            top_k=50,
+            top_p=0.95,
+            temperature=0.8,
+            pad_token_id=tokenizer.eos_token_id 
+        )
+        
+        all_completions = []
+        for sequence in output:
+            generated_text = tokenizer.decode(sequence, skip_special_tokens=True)
+            
+            # 確保內容以用戶的輸入為開頭
+            if generated_text.startswith(input_text):
+                all_completions.append(generated_text)
+            
+        # 移除重複的結果並按長度排序
+        unique_completions = sorted(list(set(all_completions)), key=len, reverse=True)
+
+        if not unique_completions:
+             # 如果模型沒有生成任何有效的補全，則返回用戶輸入本身
+             return {"completions": [input_text]}
+
+        # 返回所有唯一的補全結果 (最多 3 個)
+        return {"completions": unique_completions}
+        
+    except Exception as e:
+        print(f"推論過程發生錯誤: {e}")
+        raise HTTPException(status_code=500, detail=f"模型推論失敗：{str(e)[:50]}...")
+
+# 運行伺服器
+if __name__ == "__main__":
+    import uvicorn
+    # host 0.0.0.0 允許外部訪問，port 8000 與前端設定一致
+    uvicorn.run("api_server:app", host="0.0.0.0", port=8000, reload=True)

requirements (1).txt

@@ -0,0 +1,6 @@
+# requirements.txt
+fastapi
+uvicorn
+torch
+transformers
+pydantic