Update app.py

app.py

@@ -5,13 +5,11 @@ from PyPDF2 import PdfReader
 import google.generativeai as genai
 import os
 from langsmith import Client
-from ragas.metrics import (
-    faithfulness, answer_relevancy, context_relevancy
-)
+from ragas.metrics import faithfulness, answer_relevancy, context_relevancy
 
 # 加載模型
 openelm_model = AutoModelForCausalLM.from_pretrained("apple/OpenELM-270M", trust_remote_code=True)
-openelm_tokenizer = AutoTokenizer.from_pretrained("NousResearch/Llama-2-7b-hf")
+openelm_tokenizer = AutoTokenizer.from_pretrained("apple/OpenELM-270M")
 
 # Gemini API 設置
 GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY")
@@ -36,10 +34,6 @@ def gemini_generate(prompt, max_tokens):
     response = model.generate_content(prompt, max_output_tokens=max_tokens)
     return response.text
 
-def nvidia_generate(prompt, max_tokens):
-    # 這裡需要實現 Nvidia API 調用
-    return "Nvidia API 尚未實現"
-
 def openelm_generate(prompt, max_tokens):
     tokenized_prompt = openelm_tokenizer(prompt, return_tensors="pt")
     output_ids = openelm_model.generate(
@@ -50,41 +44,41 @@ def openelm_generate(prompt, max_tokens):
     return openelm_tokenizer.decode(output_ids[0], skip_special_tokens=True)
 
 def evaluate_response(response, context, query):
-    # 使用 RAGAS 評估回答
     faith_score = faithfulness.score([response], [context], [query])
     ans_rel_score = answer_relevancy.score([response], [query])
     ctx_rel_score = context_relevancy.score([response], [context], [query])
     return faith_score, ans_rel_score, ctx_rel_score
 
 def process_query(pdf_file, llm_choice, query, max_tokens, api_key):
-    global GOOGLE_API_KEY
-    if api_key:
-        GOOGLE_API_KEY = api_key
-        genai.configure(api_key=GOOGLE_API_KEY)
-
-    # 從 PDF 提取文本
-    pdf_path = pdf_file.name
-    context = extract_text_from_pdf(pdf_path)
-    
-    # 根據選擇的 LLM 生成回應
-    if llm_choice == "Gemini":
-        response = gemini_generate(f"上下文: {context}\n問題: {query}", max_tokens)
-    elif llm_choice == "Nvidia":
-        response = nvidia_generate(f"上下文: {context}\n問題: {query}", max_tokens)
-    else:  # OpenELM
-        response = openelm_generate(f"上下文: {context}\n問題: {query}", max_tokens)
-    
-    # 評估回應
-    faith_score, ans_rel_score, ctx_rel_score = evaluate_response(response, context, query)
-    
-    return response, faith_score, ans_rel_score, ctx_rel_score
+    try:
+        global GOOGLE_API_KEY
+        if api_key:
+            GOOGLE_API_KEY = api_key
+            genai.configure(api_key=GOOGLE_API_KEY)
+        
+        # 從 PDF 提取文本
+        pdf_path = pdf_file.name
+        context = extract_text_from_pdf(pdf_path)
+        
+        # 根據選擇的 LLM 生成回應
+        if llm_choice == "Gemini":
+            response = gemini_generate(f"上下文: {context}\n問題: {query}", max_tokens)
+        else:  # OpenELM
+            response = openelm_generate(f"上下文: {context}\n問題: {query}", max_tokens)
+        
+        # 評估回應
+        faith_score, ans_rel_score, ctx_rel_score = evaluate_response(response, context, query)
+        
+        return response, faith_score, ans_rel_score, ctx_rel_score
+    except Exception as e:
+        return str(e), 0, 0, 0  # 返回錯誤消息和零分數
 
 # Gradio 介面
 iface = gr.Interface(
     fn=process_query,
     inputs=[
         gr.File(label="上傳 PDF"),
-        gr.Dropdown(["Gemini", "Nvidia", "OpenELM"], label="選擇 LLM"),
+        gr.Dropdown(["Gemini", "OpenELM"], label="選擇 LLM"),
         gr.Textbox(label="輸入您的問題"),
         gr.Slider(minimum=50, maximum=1000, step=50, label="最大令牌數"),
         gr.Textbox(label="Gemini API Key (可選)", type="password")
@@ -96,23 +90,7 @@ iface = gr.Interface(
         gr.Number(label="上下文相關性得分")
     ],
     title="多模型 LLM 查詢介面，支持 PDF 上下文",
-    description="上傳 PDF，選擇 LLM，並提出問題。回應將使用 RAGAS 指標進行評估。",
-    css="""
-        #dev-info {
-            font-size: 0.8rem;
-            color: #888;
-            margin-top: 1rem;
-            text-align: center;
-        }
-        .gr-input text {
-            padding: 10px;
-            border-radius: 5px;
-            font-size: 1rem;
-        }
-        .gr-output.gr-slider label {
-            font-weight: bold;
-        }
-    """
+    description="上傳 PDF，選擇 LLM，並提出問題。回應將使用 RAGAS 指標進行評估。"
 )
 
 if __name__ == "__main__":