explict init cuda context

Author: shihaobai

lightllm/utils/vllm_utils.py

@@ -3,11 +3,14 @@
 
 logger = init_logger(__name__)
 try:
+    if not torch.cuda.is_initialized():
+        torch.cuda.init()
     from vllm import _custom_ops as ops
 
     vllm_ops = ops
     HAS_VLLM = True
     cutlass_scaled_mm = torch.ops._C.cutlass_scaled_mm
+
 except:
     HAS_VLLM = False
     cutlass_scaled_mm = None