python3Packages.vllm: 0.7.2 -> 0.7.3 (#384008)

Author: prusnak

pkgs/development/python-modules/vllm/0003-propagate-pythonpath.patch

@@ -1,12 +1,12 @@
 diff --git a/vllm/model_executor/models/registry.py b/vllm/model_executor/models/registry.py
-index f5a02a5b..e830f987 100644
+index 81623def..2a6e2c92 100644
 --- a/vllm/model_executor/models/registry.py
 +++ b/vllm/model_executor/models/registry.py
-@@ -482,6 +482,7 @@ def _run_in_subprocess(fn: Callable[[], _T]) -> _T:
-         returned = subprocess.run(
-             [sys.executable, "-m", "vllm.model_executor.models.registry"],
-             input=input_bytes,
-+            env={'PYTHONPATH': ':'.join(sys.path)},
-             capture_output=True)
+@@ -521,6 +521,7 @@ def _run_in_subprocess(fn: Callable[[], _T]) -> _T:
+         # contains relative imports
+         returned = subprocess.run(_SUBPROCESS_COMMAND,
+                                   input=input_bytes,
++                                  env={'PYTHONPATH': ':'.join(sys.path)},
+                                   capture_output=True)
 
          # check if the subprocess is successful

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

@@ -89,15 +89,16 @@ let
 
   vllm-flash-attn = stdenv.mkDerivation rec {
     pname = "vllm-flash-attn";
-    version = "2.6.2";
+    # https://github.com/vllm-project/flash-attention/blob/${src.rev}/vllm_flash_attn/__init__.py
+    version = "2.7.2.post1";
 
     # see CMakeLists.txt, grepping for GIT_TAG near vllm-flash-attn
     # https://github.com/vllm-project/vllm/blob/${version}/CMakeLists.txt
     src = fetchFromGitHub {
       owner = "vllm-project";
       repo = "flash-attention";
-      rev = "d4e09037abf588af1ec47d0e966b237ee376876c";
-      hash = "sha256-KFEsZlrwvCgvPzQ/pCLWcnbGq89mWE3yTDdtJSV9MII=";
+      rev = "720c94869cf2e0ff5a706e9c7f1dce0939686ade";
+      hash = "sha256-UXbBfzBCOBjRRAAOhIzt0E27VjC6xu4G1CkZfu9LKRs=";
     };
 
     dontConfigure = true;
@@ -195,7 +196,7 @@ in
 
 buildPythonPackage rec {
   pname = "vllm";
-  version = "0.7.2";
+  version = "0.7.3";
   pyproject = true;
 
   stdenv = if cudaSupport then cudaPackages.backendStdenv else args.stdenv;
@@ -204,7 +205,7 @@ buildPythonPackage rec {
     owner = "vllm-project";
     repo = pname;
     tag = "v${version}";
-    hash = "sha256-j59DpNuO5TgGD6UVGzueSTumd7mDMB4l1QytV3rFIJE=";
+    hash = "sha256-gudlikAjwZNkniKRPJYm7beoti8eHp5LaRV2/UNEibo=";
   };
 
   patches = [