Use module-level Blackwell flag in compressor (works during torch.compile)

vllm/patches/layers/deepseek_compressor.py

@@ -15,6 +15,15 @@ from vllm.model_executor.layers.linear import (
     MergedColumnParallelLinear,
 )
 from vllm.platforms import current_platform
+
+# Check at module load time if we're on Blackwell
+_IS_BLACKWELL = False
+try:
+    _cap = current_platform.get_device_capability()
+    if _cap is not None and _cap.major >= 10:
+        _IS_BLACKWELL = True
+except Exception:
+    pass
 from vllm.triton_utils import tl, triton
 from vllm.v1.attention.backend import (
     AttentionBackend,
@@ -343,8 +352,7 @@ class DeepseekCompressor(nn.Module):
         # 2. Our Blackwell attention path handles everything separately
         # Instead, we just save the state (done above) and let the attention
         # path handle compression + RoPE + cache write + attention.
-        cap = current_platform.get_device_capability()
-        if cap is not None and cap.major >= 10:
+        if _IS_BLACKWELL:
             # Blackwell: state is already saved, skip fused kernel
             return