vllm: set rocm env in wrapper

Vllm also wants bash for aiter

Author: CertainLach

pkgs/development/python-modules/vllm/default.nix

@@ -93,6 +93,9 @@
   cupy,
   flashinfer,
   nvidia-ml-py,
+  # rocm-only
+  pybind11,
+  bash,
 
   # optional-dependencies
   # audio
@@ -584,11 +587,22 @@ buildPythonPackage.override { stdenv = torch.stdenv; } (finalAttrs: {
   pythonRelaxDeps = true;
 
   pythonImportsCheck = [ "vllm" ];
-  makeWrapperArgs = lib.optionals cudaSupport [
-    "--set"
-    "VLLM_NCCL_SO_PATH"
-    "${cudaPackages.nccl}/lib/libnccl.so"
-  ];
+  makeWrapperArgs =
+    lib.optionals cudaSupport [
+      "--set"
+      "VLLM_NCCL_SO_PATH"
+      "${cudaPackages.nccl}/lib/libnccl.so"
+    ]
+    ++ lib.optionals rocmSupport [
+      "--set"
+      "HIP_DEVICE_LIB_PATH"
+      "${rocmPackages.rocm-device-libs}/amdgcn/bitcode"
+
+      "--prefix"
+      "PATH"
+      ":"
+      "${rocmPackages.clr}/bin:${bash}/bin"
+    ];
 
   passthru = {
     # make internal dependency available to overlays