fix: use dtype= instead of deprecated torch_dtype=

server.py

@@ -151,7 +151,7 @@ def generate_streaming(prompt: str, max_tokens: int = 200,
 print(f"Loading base model: {BASE_MODEL_REPO} ...")
 BASE_MODEL = AutoModelForCausalLM.from_pretrained(
     BASE_MODEL_REPO,
-    torch_dtype=torch.float32,
+    dtype=torch.float32,
     device_map="cpu",
     low_cpu_mem_usage=True,
 )