feat: use fa3 mla by default on hopper (#5210)

Co-authored-by: yundai424 <yundai424@gmail.com>
Co-authored-by: hebiao064 <hebiaobuaa@gmail.com>

python/sglang/srt/utils.py

@@ -1828,3 +1828,12 @@ def fast_topk(values, topk, dim):
     else:
         # Use topk for efficiency with larger k values
         return torch.topk(values, topk, dim=dim)
+
+
+def is_hopper_with_cuda_12_3():
+    if not is_cuda():
+        return False
+    is_hopper = torch.cuda.get_device_capability()[0] == 9
+    cuda_version = torch.version.cuda.split(".")
+    is_cuda_compatible = int(cuda_version[0]) == 12 and int(cuda_version[1]) >= 3
+    return is_hopper and is_cuda_compatible