[Bug] DeepGemm: Fix Cuda Init Error

vllm/utils/deep_gemm.py

@@ -45,30 +45,34 @@ def _resolve_symbol(module, new: str, old: str) -> Callable[..., Any] | None:
     return None
 
 
-if not has_deep_gemm():
-    _fp8_gemm_nt_impl: Callable[..., Any] | None = None
-    _grouped_impl: Callable[..., Any] | None = None
-    _grouped_masked_impl: Callable[..., Any] | None = None
-    _per_block_cast_impl: Callable[..., Any] | None = None
-else:
-    _dg = importlib.import_module("deep_gemm")  # type: ignore
-
-    _fp8_gemm_nt_impl = _resolve_symbol(
-        _dg,
-        "fp8_gemm_nt",
-        "gemm_fp8_fp8_bf16_nt",
-    )
+_fp8_gemm_nt_impl: Callable[..., Any] | None = None
+_grouped_impl: Callable[..., Any] | None = None
+_grouped_masked_impl: Callable[..., Any] | None = None
+_per_block_cast_impl: Callable[..., Any] | None = None
+
+
+def _lazy_init() -> None:
+    """Import deep_gemm and resolve symbols on first use."""
+    global _fp8_gemm_nt_impl, _grouped_impl, _grouped_masked_impl
+
+    # fast path
+    if (_fp8_gemm_nt_impl is not None or _grouped_impl is not None
+            or _grouped_masked_impl is not None):
+        return
+
+    if not has_deep_gemm():
+        return
+
+    _dg = importlib.import_module("deep_gemm")
+
+    _fp8_gemm_nt_impl = _resolve_symbol(_dg, "fp8_gemm_nt",
+                                        "gemm_fp8_fp8_bf16_nt")
     _grouped_impl = _resolve_symbol(
-        _dg,
-        "m_grouped_fp8_gemm_nt_contiguous",
-        "m_grouped_gemm_fp8_fp8_bf16_nt_contiguous",
-    )
+        _dg, "m_grouped_fp8_gemm_nt_contiguous",
+        "m_grouped_gemm_fp8_fp8_bf16_nt_contiguous")
     _grouped_masked_impl = _resolve_symbol(
-        _dg,
-        "fp8_m_grouped_gemm_nt_masked",
-        "m_grouped_gemm_fp8_fp8_bf16_nt_masked",
-    )
-
+        _dg, "fp8_m_grouped_gemm_nt_masked",
+        "m_grouped_gemm_fp8_fp8_bf16_nt_masked")
     # Try to get per_token_cast_to_fp8 from DeepGEMM math utils.
     try:
         _math_mod = importlib.import_module(
@@ -80,24 +84,28 @@ def _resolve_symbol(module, new: str, old: str) -> Callable[..., Any] | None:
 
 
 def fp8_gemm_nt(*args, **kwargs):
+    _lazy_init()
     if _fp8_gemm_nt_impl is None:
         return _missing(*args, **kwargs)
     return _fp8_gemm_nt_impl(*args, **kwargs)
 
 
 def m_grouped_fp8_gemm_nt_contiguous(*args, **kwargs):
+    _lazy_init()
     if _grouped_impl is None:
         return _missing(*args, **kwargs)
     return _grouped_impl(*args, **kwargs)
 
 
 def fp8_m_grouped_gemm_nt_masked(*args, **kwargs):
+    _lazy_init()
     if _grouped_masked_impl is None:
         return _missing(*args, **kwargs)
     return _grouped_masked_impl(*args, **kwargs)
 
 
 def per_block_cast_to_fp8(x, *args, **kwargs):
+    _lazy_init()
     if _per_block_cast_impl is not None and is_blackwell_deep_gemm_used():
         return _per_block_cast_impl(x, use_ue8m0=True)
     # TODO: refactor the `per_block_cast_to_fp8` from tests to vllm utils