Set CUDA minimum versions for hopper/blackwell cutlass kernels

build.toml

@@ -127,6 +127,7 @@ src = [
 [kernel.cutlass_w8a8_hopper]
 backend = "cuda"
 cuda-capabilities = ["9.0a"]
+cuda-minver = "12.0"
 depends = [
     "cutlass_3_9",
     "torch",
@@ -167,6 +168,7 @@ cuda-capabilities = [
     "10.1a",
     "12.0a",
 ]
+cuda-minver = "12.9"
 depends = [
     "cutlass_3_9",
     "torch",

flake.lock

@@ -98,15 +98,16 @@
         ]
       },
       "locked": {
-        "lastModified": 1751014803,
-        "narHash": "sha256-9Xfq2k3uPfB602NwQF+zAY2GQZiKUN1G7Q6XiDCUR8Y=",
+        "lastModified": 1751544642,
+        "narHash": "sha256-WZC/9SO3a92pH5AjvPn3GXSdupysFI6JglF+t4o/TYQ=",
         "owner": "huggingface",
         "repo": "kernel-builder",
-        "rev": "bbc4e712ff2046e217818e97de2201e2b996756e",
+        "rev": "e12cd4ae24b7315af65db72483e6578909f3dac8",
         "type": "github"
       },
       "original": {
         "owner": "huggingface",
+        "ref": "kernel-cuda-minver",
         "repo": "kernel-builder",
         "type": "github"
       }

flake.nix

@@ -2,7 +2,7 @@
   description = "Flake for quantization kernels";
 
   inputs = {
-    kernel-builder.url = "github:huggingface/kernel-builder";
+    kernel-builder.url = "github:huggingface/kernel-builder/kernel-cuda-minver";
   };
 
   outputs =