[AMD] support gfx1250 on ROCM 10 (#36871)

Co-authored-by: HAI <hixiao@gmail.com>
Co-authored-by: Kao <akao@amd.com>
Co-authored-by: wunhuang <wunhuang@amd.com>
Co-authored-by: Thomas Wang <1am9trash@gmail.com>
Co-authored-by: Xinyi Song <86638975+RolaoDenthu@users.noreply.github.com>
Co-authored-by: Lin, Soga <soga.lin@amd.com>
Co-authored-by: kk <43161300+kkHuang-amd@users.noreply.github.com>
Co-authored-by: Bingxu Chen <bingxche@amd.com>
Co-authored-by: sogalin_codegen <39478626+sogalin@users.noreply.github.com>
Co-authored-by: Thomas Wang <thomawan@amd.com>

python/sglang/srt/model_executor/runner_utils/capture_mode.py

@@ -28,6 +28,7 @@ from sglang.srt.model_executor.runner_backend_utils.breakable_cuda_graph.context
     is_in_breakable_cuda_graph,
 )
 from sglang.srt.runtime_context import get_flags
+from sglang.srt.utils import is_gfx1250_supported
 
 # Detect whether the current forward pass is in capture mode.
 is_capture_mode = False
@@ -56,7 +57,7 @@ def compile_in_capture_mode(func):
     torch.compile during cuda-graph capture without paying the
     compilation cost in the eager forward path.
     """
-    if is_capture_mode:
+    if is_capture_mode and not is_gfx1250_supported():
         return torch.compile(func)
     return func