Merge

app.py

@@ -9,21 +9,15 @@ MODEL_NAMES = {
 }
 
 HF_TOKEN = os.getenv("HF_TOKEN")
-<<<<<<< HEAD
 
 
 def load_model(model_path):
     tokenizer = AutoTokenizer.from_pretrained(
         model_path, trust_remote_code=True, token=HF_TOKEN
     )
-    model = AutoModelForCausalLM.from_pretrained(
-        model_path, torch_dtype=torch.float16, trust_remote_code=True, token=HF_TOKEN
-    ).cuda()
-=======
-
-def load_model(model_path):
-    tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True, token=HF_TOKEN)
-    config = AutoConfig.from_pretrained(model_path, trust_remote_code=True, token=HF_TOKEN)
+    config = AutoConfig.from_pretrained(
+        model_path, trust_remote_code=True, token=HF_TOKEN
+    )
     if hasattr(config, "quantization_config"):
         del config.quantization_config  # 刪除量化配置，避免使用 FP8
 
@@ -35,13 +29,14 @@ def load_model(model_path):
         torch_dtype=torch.float16,
         device_map="auto",
     )
->>>>>>> c4a529954b9cc11b7b8c2dcdac51f3ef3a83c616
     return model, tokenizer
 
+
 # 初始化預設模型
 current_model_name = "DeepSeek-R1-Distill-Llama-8B"
 current_model, current_tokenizer = load_model(MODEL_NAMES[current_model_name])
 
+
 def chat(message, history, model_name):
     global current_model, current_tokenizer, current_model_name
 
@@ -57,6 +52,7 @@ def chat(message, history, model_name):
 
     return response
 
+
 with gr.Blocks() as app:
     gr.Markdown("## Chatbot with DeepSeek Models")
 
@@ -73,6 +69,8 @@ with gr.Blocks() as app:
         save_history=True,
     )
 
-    model_selector.change(fn=lambda model_name: None, inputs=[model_selector], outputs=[])
+    model_selector.change(
+        fn=lambda model_name: None, inputs=[model_selector], outputs=[]
+    )
 
 app.launch()

requirements.txt

@@ -1,8 +1,3 @@
-<<<<<<< HEAD
-transformers>=4.37.2
-torch>=2.1.0
-=======
 transformers>=4.38.0
 torch>=2.1.0
 accelerate
->>>>>>> c4a529954b9cc11b7b8c2dcdac51f3ef3a83c616