diff --git a/pkgs/development/python-modules/kserve/default.nix b/pkgs/development/python-modules/kserve/default.nix
index bd7125cffe1af..04cde32164969 100644
--- a/pkgs/development/python-modules/kserve/default.nix
+++ b/pkgs/development/python-modules/kserve/default.nix
@@ -3,6 +3,7 @@
   stdenv,
   buildPythonPackage,
   fetchFromGitHub,
+  fetchpatch2,
 
   # build-system
   setuptools,
@@ -51,7 +52,7 @@
   tomlkit,
 }:
 
-buildPythonPackage rec {
+buildPythonPackage (finalAttrs: {
   pname = "kserve";
   version = "0.16.0";
   pyproject = true;
@@ -59,23 +60,22 @@ buildPythonPackage rec {
   src = fetchFromGitHub {
     owner = "kserve";
     repo = "kserve";
-    tag = "v${version}";
+    tag = "v${finalAttrs.version}";
     hash = "sha256-f6ILZMLxfckEpy7wSgCqUx89JWSnn0DbQiqRSHcQHms=";
   };
 
-  # Fix vllm 0.12.0 compatibility
-  # Patch submitted upstream: https://github.com/kserve/kserve/pull/4882
-  postPatch = ''
-    substituteInPlace kserve/protocol/rest/openai/types/__init__.py \
-      --replace-fail \
-        "from vllm.entrypoints.openai.protocol import EmbeddingRequest, EmbeddingResponse as Embedding, EmbeddingResponseData, EmbeddingCompletionRequest" \
-        "from vllm.entrypoints.pooling.embed.protocol import EmbeddingRequest, EmbeddingResponse as Embedding, EmbeddingResponseData, EmbeddingCompletionRequest" \
-      --replace-fail \
-        "from vllm.entrypoints.openai.protocol import RerankRequest, RerankResponse as Rerank" \
-        "from vllm.entrypoints.pooling.score.protocol import RerankRequest, RerankResponse as Rerank"
-  '';
+  patches = [
+    # Fix vllm imports in python/kserve/kserve/protocol/rest/openai/types/__init__.py
+    # Submitted upstream: https://github.com/kserve/kserve/pull/4882
+    (fetchpatch2 {
+      name = "update-vllm-imports-to-fix-compat";
+      url = "https://github.com/kserve/kserve/commit/dd1575501e56f588103f448efca684bc54569b81.patch";
+      stripLen = 2;
+      hash = "sha256-K0ImsDADhH6G3R+27nRX/sD7UdRXptYIkLaoxuwB8+M=";
+    })
+  ];
 
-  sourceRoot = "${src.name}/python/kserve";
+  sourceRoot = "${finalAttrs.src.name}/python/kserve";
 
   pythonRelaxDeps = [
     "fastapi"
@@ -142,7 +142,7 @@ buildPythonPackage rec {
     pytestCheckHook
     tomlkit
   ]
-  ++ lib.concatAttrValues optional-dependencies;
+  ++ lib.concatAttrValues finalAttrs.passthru.optional-dependencies;
 
   pythonImportsCheck = [ "kserve" ];
 
@@ -205,8 +205,8 @@ buildPythonPackage rec {
   meta = {
     description = "Standardized Serverless ML Inference Platform on Kubernetes";
     homepage = "https://github.com/kserve/kserve/tree/master/python/kserve";
-    changelog = "https://github.com/kserve/kserve/releases/tag/${src.tag}";
+    changelog = "https://github.com/kserve/kserve/releases/tag/${finalAttrs.src.tag}";
     license = lib.licenses.asl20;
     maintainers = with lib.maintainers; [ GaetanLepage ];
   };
-}
+})
diff --git a/pkgs/development/python-modules/vllm/default.nix b/pkgs/development/python-modules/vllm/default.nix
index 86b3c493f3357..9cf7eaca36a06 100644
--- a/pkgs/development/python-modules/vllm/default.nix
+++ b/pkgs/development/python-modules/vllm/default.nix
@@ -133,8 +133,8 @@ let
   cutlass-flashmla = fetchFromGitHub {
     owner = "NVIDIA";
     repo = "cutlass";
-    tag = "v3.9.0";
-    hash = "sha256-Q6y/Z6vahASeSsfxvZDwbMFHGx8CnsF90IlveeVLO9g=";
+    rev = "147f5673d0c1c3dcf66f78d677fd647e4a020219";
+    hash = "sha256-dHQto08IwTDOIuFUp9jwm1MWkFi8v2YJ/UESrLuG71g=";
   };
 
   flashmla = stdenv.mkDerivation {
@@ -148,8 +148,8 @@ let
       name = "FlashMLA-source";
       owner = "vllm-project";
       repo = "FlashMLA";
-      rev = "46d64a8ebef03fa50b4ae74937276a5c940e3f95";
-      hash = "sha256-jtMzWB5hKz8mJGsdK6q4YpQbGp9IrQxbwmB3a64DIl0=";
+      rev = "c2afa9cb93e674d5a9120a170a6da57b89267208";
+      hash = "sha256-pKlwxV6G9iHag/jbu3bAyvYvnu5TbrQwUMFV0AlGC3s=";
     };
 
     dontConfigure = true;
@@ -324,14 +324,14 @@ in
 
 buildPythonPackage.override { stdenv = torch.stdenv; } (finalAttrs: {
   pname = "vllm";
-  version = "0.14.0";
+  version = "0.15.1";
   pyproject = true;
 
   src = fetchFromGitHub {
     owner = "vllm-project";
     repo = "vllm";
     tag = "v${finalAttrs.version}";
-    hash = "sha256-gUfEjoNgS/FgSDqQDnDe/onWGigzwkKuPgmdRZHVGn0=";
+    hash = "sha256-qsAvcOB8ugGlBqBrLfNHqaIUcxLwaXBTg8xWRnGyd94=";
   };
 
   patches = [