dynamic distpatch of fp8 kernels (#14245)

Signed-off-by: Jeff Daily <jeff.daily@amd.com>

vllm/platforms/rocm.py

@@ -231,3 +231,20 @@ class RocmPlatform(Platform):
     @classmethod
     def get_device_communicator_cls(cls) -> str:
         return "vllm.distributed.device_communicators.cuda_communicator.CudaCommunicator"  # noqa
+
+    @classmethod
+    def supports_fp8(cls) -> bool:
+        gcn_arch = torch.cuda.get_device_properties(0).gcnArchName
+        return any(gfx in gcn_arch for gfx in ['gfx94', 'gfx95', 'gfx12'])
+
+    @classmethod
+    def is_fp8_fnuz(cls) -> bool:
+        # only device 0 is checked, this assumes MI300 platforms are homogeneous
+        return 'gfx94' in torch.cuda.get_device_properties(0).gcnArchName
+
+    @classmethod
+    def fp8_dtype(cls) -> torch.dtype:
+        if cls.is_fp8_fnuz():
+            return torch.float8_e4m3fnuz
+        else:
+            return torch.float8_e4m3fn