diff --git a/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0-foss-2023b.eb b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0-foss-2023b.eb
new file mode 100644
index 000000000000..ca28c8dcea6f
--- /dev/null
+++ b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0-foss-2023b.eb
@@ -0,0 +1,191 @@
+name = 'PyTorch'
+version = '2.3.0'
+
+homepage = 'https://pytorch.org/'
+description = """Tensors and Dynamic neural networks in Python with strong GPU acceleration.
+PyTorch is a deep learning framework that puts Python first."""
+
+toolchain = {'name': 'foss', 'version': '2023b'}
+
+source_urls = [GITHUB_RELEASE]
+sources = ['%(namelower)s-v%(version)s.tar.gz']
+patches = [
+    'PyTorch-1.7.0_disable-dev-shm-test.patch',
+    'PyTorch-1.12.1_add-hypothesis-suppression.patch',
+    'PyTorch-1.12.1_fix-test_cpp_extensions_jit.patch',
+    'PyTorch-1.12.1_fix-TestTorch.test_to.patch',
+    'PyTorch-1.12.1_skip-test_round_robin.patch',
+    'PyTorch-1.13.1_fix-gcc-12-warning-in-fbgemm.patch',
+    'PyTorch-1.13.1_fix-protobuf-dependency.patch',
+    'PyTorch-1.13.1_fix-warning-in-test-cpp-api.patch',
+    'PyTorch-1.13.1_skip-failing-singular-grad-test.patch',
+    'PyTorch-1.13.1_skip-tests-without-fbgemm.patch',
+    'PyTorch-2.0.1_avoid-test_quantization-failures.patch',
+    'PyTorch-2.0.1_fix-skip-decorators.patch',
+    'PyTorch-2.0.1_fix-vsx-loadu.patch',
+    'PyTorch-2.0.1_skip-failing-gradtest.patch',
+    'PyTorch-2.0.1_skip-test_shuffle_reproducibility.patch',
+    'PyTorch-2.0.1_skip-tests-skipped-in-subprocess.patch',
+    'PyTorch-2.1.0_fix-vsx-vector-shift-functions.patch',
+    'PyTorch-2.1.0_increase-tolerance-functorch-test_vmapvjpvjp.patch',
+    'PyTorch-2.1.0_remove-test-requiring-online-access.patch',
+    'PyTorch-2.1.0_skip-diff-test-on-ppc.patch',
+    'PyTorch-2.1.0_skip-dynamo-test_predispatch.patch',
+    'PyTorch-2.1.0_skip-test_jvp_linalg_det_singular.patch',
+    'PyTorch-2.1.2_skip-cpu_repro-test-without-vectorization.patch',
+    'PyTorch-2.1.2_workaround_dynamo_failure_without_nnpack.patch',
+    'PyTorch-2.3.0_disable_test_linear_package_if_no_half_types_are_available.patch',
+    'PyTorch-2.3.0_disable_DataType_dependent_test_if_tensorboard_is_not_available.patch',
+    'PyTorch-2.3.0_skip-test_init_from_local_shards.patch',
+    'PyTorch-2.3.0_no-cuda-stubs-rpath.patch',
+    'PyTorch-2.3.0_disable-gcc12-warning.patch',
+    'PyTorch-2.3.0_fix-test_extension_backend-without-vectorization.patch',
+    'PyTorch-2.3.0_disable_tests_which_need_network_download.patch',
+    'PyTorch-2.3.0_avoid_caffe2_test_cpp_jit.patch',
+    'PyTorch-2.3.0_fix_missing_masked_load_for_int_type.patch',
+    'PyTorch-2.3.0_skip_test_var_mean_differentiable.patch',
+    'PyTorch-2.3.0_skip_test_sdpa_nn_functional_scaled_dot_product_attention_cpu.patch',
+    'PyTorch-2.3.0_fix-mkldnn-avx512-f32-bias.patch',
+]
+checksums = [
+    {'pytorch-v2.3.0.tar.gz': '69579513b26261bbab32e13b7efc99ad287fcf3103087f2d4fdf1adacd25316f'},
+    {'PyTorch-1.7.0_disable-dev-shm-test.patch': '622cb1eaeadc06e13128a862d9946bcc1f1edd3d02b259c56a9aecc4d5406b8a'},
+    {'PyTorch-1.12.1_add-hypothesis-suppression.patch':
+     'e71ffb94ebe69f580fa70e0de84017058325fdff944866d6bd03463626edc32c'},
+    {'PyTorch-1.12.1_fix-test_cpp_extensions_jit.patch':
+     '1efc9850c431d702e9117d4766277d3f88c5c8b3870997c9974971bce7f2ab83'},
+    {'PyTorch-1.12.1_fix-TestTorch.test_to.patch': '75f27987c3f25c501e719bd2b1c70a029ae0ee28514a97fe447516aee02b1535'},
+    {'PyTorch-1.12.1_skip-test_round_robin.patch': '63d4849b78605aa088fdff695637d9473ea60dee603a3ff7f788690d70c55349'},
+    {'PyTorch-1.13.1_fix-gcc-12-warning-in-fbgemm.patch':
+     '5c7be91a6096083a0b1315efe0001537499c600f1f569953c6a2c7f4cc1d0910'},
+    {'PyTorch-1.13.1_fix-protobuf-dependency.patch':
+     '8bd755a0cab7233a243bc65ca57c9630dfccdc9bf8c9792f0de4e07a644fcb00'},
+    {'PyTorch-1.13.1_fix-warning-in-test-cpp-api.patch':
+     'bdde0f2105215c95a54de64ec4b1a4520528510663174fef6d5b900eb1db3937'},
+    {'PyTorch-1.13.1_skip-failing-singular-grad-test.patch':
+     '72688a57b2bb617665ad1a1d5e362c5111ae912c10936bb38a089c0204729f48'},
+    {'PyTorch-1.13.1_skip-tests-without-fbgemm.patch':
+     '481e595f673baf8ae58b41697a6792b83048b0264aa79b422f48cd8c22948bb7'},
+    {'PyTorch-2.0.1_avoid-test_quantization-failures.patch':
+     '02e3f47e4ed1d7d6077e26f1ae50073dc2b20426269930b505f4aefe5d2f33cd'},
+    {'PyTorch-2.0.1_fix-skip-decorators.patch': '2039012cef45446065e1a2097839fe20bb29fe3c1dcc926c3695ebf29832e920'},
+    {'PyTorch-2.0.1_fix-vsx-loadu.patch': 'a0ffa61da2d47c6acd09aaf6d4791e527d8919a6f4f1aa7ed38454cdcadb1f72'},
+    {'PyTorch-2.0.1_skip-failing-gradtest.patch': '8030bdec6ba49b057ab232d19a7f1a5e542e47e2ec340653a246ec9ed59f8bc1'},
+    {'PyTorch-2.0.1_skip-test_shuffle_reproducibility.patch':
+     '7047862abc1abaff62954da59700f36d4f39fcf83167a638183b1b7f8fec78ae'},
+    {'PyTorch-2.0.1_skip-tests-skipped-in-subprocess.patch':
+     '166c134573a95230e39b9ea09ece3ad8072f39d370c9a88fb2a1e24f6aaac2b5'},
+    {'PyTorch-2.1.0_fix-vsx-vector-shift-functions.patch':
+     '3793b4b878be1abe7791efcbd534774b87862cfe7dc4774ca8729b6cabb39e7e'},
+    {'PyTorch-2.1.0_increase-tolerance-functorch-test_vmapvjpvjp.patch':
+     'aef38adf1210d0c5455e91d7c7a9d9e5caad3ae568301e0ba9fc204309438e7b'},
+    {'PyTorch-2.1.0_remove-test-requiring-online-access.patch':
+     '35184b8c5a1b10f79e511cc25db3b8a5585a5d58b5d1aa25dd3d250200b14fd7'},
+    {'PyTorch-2.1.0_skip-diff-test-on-ppc.patch': '394157dbe565ffcbc1821cd63d05930957412156cc01e949ef3d3524176a1dda'},
+    {'PyTorch-2.1.0_skip-dynamo-test_predispatch.patch':
+     '6298daf9ddaa8542850eee9ea005f28594ab65b1f87af43d8aeca1579a8c4354'},
+    {'PyTorch-2.1.0_skip-test_jvp_linalg_det_singular.patch':
+     '5229ca88a71db7667a90ddc0b809b2c817698bd6e9c5aaabd73d3173cf9b99fe'},
+    {'PyTorch-2.1.2_skip-cpu_repro-test-without-vectorization.patch':
+     '7ace835af60c58d9e0754a34c19d4b9a0c3a531f19e5d0eba8e2e49206eaa7eb'},
+    {'PyTorch-2.1.2_workaround_dynamo_failure_without_nnpack.patch':
+     'fb96eefabf394617bbb3fbd3a7a7c1aa5991b3836edc2e5d2a30e708bfe49ba1'},
+    {'PyTorch-2.3.0_disable_test_linear_package_if_no_half_types_are_available.patch':
+     '23416f2d9d5226695ec3fbea0671e3650c655c19deefd3f0f8ddab5afa50f485'},
+    {'PyTorch-2.3.0_disable_DataType_dependent_test_if_tensorboard_is_not_available.patch':
+     '0dcbdfde6752c3ff54c5376f521b4a742167669feb7f0f1d4e1d4d55f72b664f'},
+    {'PyTorch-2.3.0_skip-test_init_from_local_shards.patch':
+     '90ed9c2870f57ee6dc032d00873a37e2217a2b92a13035ded1c25ad5306455f2'},
+    {'PyTorch-2.3.0_no-cuda-stubs-rpath.patch':
+     '7ba26824b5def7379cff02ae821a080698e6affea0da45bc846e9ecb89939cb1'},
+    {'PyTorch-2.3.0_disable-gcc12-warning.patch':
+     'a8a624e1a2a5f4c82610173e50bd0f853e49bd5621b432f5aac689f9f6eb1514'},
+    {'PyTorch-2.3.0_fix-test_extension_backend-without-vectorization.patch':
+     '36aa2d5ba175be17f4e996f4fb2d544fe477d4a0bd0644cd59a85063779afc8e'},
+    {'PyTorch-2.3.0_disable_tests_which_need_network_download.patch':
+     'b7fd1a5135dfd4098cdc054182f7bf84a23ac98462a00477712182b5442da855'},
+    {'PyTorch-2.3.0_avoid_caffe2_test_cpp_jit.patch':
+     '041adcd91d994b8c2ab57d227f081cd57e572c157117b37171e1eb8eb576f8fc'},
+    {'PyTorch-2.3.0_fix_missing_masked_load_for_int_type.patch':
+     'aa6ff764f3f7bf84372a8a257fe1b4ae6dc4b9744ad35f0f9015f2696c62a41e'},
+    {'PyTorch-2.3.0_skip_test_var_mean_differentiable.patch':
+     '9703fd0f1fca8916f6d79d83e9a7efe8e3f717362a5fdaa8f5d9da90d0c75018'},
+    {'PyTorch-2.3.0_skip_test_sdpa_nn_functional_scaled_dot_product_attention_cpu.patch':
+     '7955f2655db3da18606574fdcbc5990be24098f49ad1db5e86ea756ea1cc506f'},
+    {'PyTorch-2.3.0_fix-mkldnn-avx512-f32-bias.patch':
+     'ee07d21c3ac7aeb0bd0e39507b18a417b9125284a529102929c4b5c6727c2976'},
+]
+
+osdependencies = [OS_PKG_IBVERBS_DEV]
+
+builddependencies = [
+    ('CMake', '3.27.6'),
+    ('hypothesis', '6.90.0'),
+    # For tests
+    ('pytest-flakefinder', '1.1.0'),
+    ('pytest-rerunfailures', '14.0'),
+    ('pytest-shard', '0.1.2'),
+    ('tlparse', '0.3.5'),
+    ('optree', '0.13.0'),
+    ('unittest-xml-reporting', '3.1.0'),
+]
+
+dependencies = [
+    ('Ninja', '1.11.1'),  # Required for JIT compilation of C++ extensions
+    ('Python', '3.11.5'),
+    ('Python-bundle-PyPI', '2023.10'),
+    ('protobuf', '25.3'),
+    ('protobuf-python', '4.25.3'),
+    ('pybind11', '2.11.1'),
+    ('SciPy-bundle', '2023.11'),
+    ('PyYAML', '6.0.1'),
+    ('MPFR', '4.2.1'),
+    ('GMP', '6.3.0'),
+    ('numactl', '2.0.16'),
+    ('FFmpeg', '6.0'),
+    ('Pillow', '10.2.0'),
+    ('expecttest', '0.2.1'),
+    ('networkx', '3.2.1'),
+    ('sympy', '1.12'),
+    ('Z3', '4.13.0',),
+]
+
+use_pip = True
+buildcmd = '%(python)s setup.py build'  # Run the (long) build in the build step
+
+excluded_tests = {
+    '': [
+        # This test seems to take too long on NVIDIA Ampere at least.
+        'distributed/test_distributed_spawn',
+        # Broken on CUDA 11.6/11.7: https://github.com/pytorch/pytorch/issues/75375
+        'distributions/test_constraints',
+        # no xdoctest
+        'doctests',
+        # failing on broadwell
+        # See https://github.com/easybuilders/easybuild-easyconfigs/issues/17712
+        'test_native_mha',
+        # intermittent failures on various systems
+        # See https://github.com/easybuilders/easybuild-easyconfigs/issues/17712
+        'distributed/rpc/test_tensorpipe_agent',
+        # This test is expected to fail when run in their CI, but won't in our case.
+        # It just checks for a "CI" env variable
+        'test_ci_sanity_check_fail',
+        # This fails consistently and is disabled upstream
+        # See https://github.com/pytorch/pytorch/issues/100152 and
+        # https://github.com/pytorch/pytorch/pull/124712
+        'test_cpp_extensions_open_device_registration',
+
+    ]
+}
+
+local_test_opts = '--continue-through-error --pipe-logs --verbose %(excluded_tests)s'
+runtest = 'cd test && PYTHONUNBUFFERED=1 %(python)s run_test.py ' + local_test_opts
+
+# Especially test_quantization has a few corner cases that are triggered by the random input values,
+# those cannot be easily avoided, see https://github.com/pytorch/pytorch/issues/107030
+# So allow a low number of tests to fail as the tests "usually" succeed
+max_failed_tests = 6
+
+tests = ['PyTorch-check-cpp-extension.py']
+
+moduleclass = 'ai'
diff --git a/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_avoid_caffe2_test_cpp_jit.patch b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_avoid_caffe2_test_cpp_jit.patch
new file mode 100644
index 000000000000..7e322c371074
--- /dev/null
+++ b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_avoid_caffe2_test_cpp_jit.patch
@@ -0,0 +1,106 @@
+commit 9e09454cdefc14397c57b2435d3da0c0413dbb8e
+Author: Ake Sandgren <ake.sandgren@hpc2n.umu.se>
+Date:   Thu Feb 27 15:18:52 2025 +0100
+
+    Avoid IS_SANDCASTLE tripping on //caffe2/test/cpp/jit:test_custom_class_registrations
+
+diff --git a/test/export/test_export.py b/test/export/test_export.py
+index 9d1006745b4..b7ee8c1d451 100644
+--- a/test/export/test_export.py
++++ b/test/export/test_export.py
+@@ -3917,7 +3917,7 @@ class TestExportCustomClass(TorchTestCase):
+     def setUp(self):
+         if IS_FBCODE:
+             lib_file_path = "//caffe2/test/cpp/jit:test_custom_class_registrations"
+-        elif IS_SANDCASTLE or IS_MACOS:
++        elif False or IS_MACOS:
+             raise unittest.SkipTest("non-portable load_library call used in test")
+         elif IS_WINDOWS:
+             lib_file_path = find_library_location('torchbind_test.dll')
+diff --git a/test/export/test_lift_unlift.py b/test/export/test_lift_unlift.py
+index edbfa0803d4..a93728f8bd3 100644
+--- a/test/export/test_lift_unlift.py
++++ b/test/export/test_lift_unlift.py
+@@ -146,7 +146,7 @@ class TestLift(TestCase):
+     def setUp(self):
+         if IS_MACOS:
+             raise unittest.SkipTest("non-portable load_library call used in test")
+-        elif IS_SANDCASTLE or IS_FBCODE:
++        elif False or IS_FBCODE:
+             torch.ops.load_library(
+                 "//caffe2/test/cpp/jit:test_custom_class_registrations"
+             )
+@@ -379,7 +379,7 @@ class ConstantAttrMapTest(TestCase):
+     def setUp(self):
+         if IS_MACOS:
+             raise unittest.SkipTest("non-portable load_library call used in test")
+-        elif IS_SANDCASTLE or IS_FBCODE:
++        elif False or IS_FBCODE:
+             torch.ops.load_library(
+                 "//caffe2/test/cpp/jit:test_custom_class_registrations"
+             )
+diff --git a/test/export/test_serialize.py b/test/export/test_serialize.py
+index 08381d95b44..59016a874ad 100644
+--- a/test/export/test_serialize.py
++++ b/test/export/test_serialize.py
+@@ -823,7 +823,7 @@ class TestSaveLoad(TestCase):
+ @unittest.skipIf(not torchdynamo.is_dynamo_supported(), "dynamo doesn't support")
+ class TestSerializeCustomClass(TestCase):
+     def setUp(self):
+-        if IS_SANDCASTLE or IS_FBCODE:
++        if False or IS_FBCODE:
+             torch.ops.load_library(
+                 "//caffe2/test/cpp/jit:test_custom_class_registrations"
+             )
+diff --git a/test/export/test_torchbind.py b/test/export/test_torchbind.py
+index 362064812d4..e5b2f4ae511 100644
+--- a/test/export/test_torchbind.py
++++ b/test/export/test_torchbind.py
+@@ -21,7 +21,7 @@ class TestExportTorchbind(TestCase):
+     def setUp(self):
+         if IS_MACOS:
+             raise unittest.SkipTest("non-portable load_library call used in test")
+-        elif IS_SANDCASTLE or IS_FBCODE:
++        elif False or IS_FBCODE:
+             torch.ops.load_library(
+                 "//caffe2/test/cpp/jit:test_custom_class_registrations"
+             )
+diff --git a/test/export/test_unflatten.py b/test/export/test_unflatten.py
+index 64f123136ba..d0baef3777c 100644
+--- a/test/export/test_unflatten.py
++++ b/test/export/test_unflatten.py
+@@ -552,7 +552,7 @@ class TestUnflatten(TestCase):
+     def test_unflatten_constant_obj(self):
+         if IS_MACOS:
+             raise unittest.SkipTest("non-portable load_library call used in test")
+-        elif IS_SANDCASTLE or IS_FBCODE:
++        elif False or IS_FBCODE:
+             torch.ops.load_library(
+                 "//caffe2/test/cpp/jit:test_custom_class_registrations"
+             )
+diff --git a/test/higher_order_ops/test_with_effects.py b/test/higher_order_ops/test_with_effects.py
+index 0c8f12a1058..3347b6ca1fc 100644
+--- a/test/higher_order_ops/test_with_effects.py
++++ b/test/higher_order_ops/test_with_effects.py
+@@ -27,7 +27,7 @@ class TestWithEffects(TestCase):
+     def setUp(self):
+         if IS_MACOS:
+             raise unittest.SkipTest("non-portable load_library call used in test")
+-        elif IS_SANDCASTLE or IS_FBCODE:
++        elif False or IS_FBCODE:
+             torch.ops.load_library(
+                 "//caffe2/test/cpp/jit:test_custom_class_registrations"
+             )
+diff --git a/test/test_weak.py b/test/test_weak.py
+index cc9ff6f58ee..54ee73a6dae 100644
+--- a/test/test_weak.py
++++ b/test/test_weak.py
+@@ -600,7 +600,7 @@ class WeakKeyDictionaryScriptObjectTestCase(TestCase):
+ 
+     def __init__(self, *args, **kw):
+         unittest.TestCase.__init__(self, *args, **kw)
+-        if IS_SANDCASTLE or IS_FBCODE:
++        if False or IS_FBCODE:
+             torch.ops.load_library(
+                 "//caffe2/test/cpp/jit:test_custom_class_registrations"
+             )
diff --git a/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_disable-gcc12-warning.patch b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_disable-gcc12-warning.patch
new file mode 100644
index 000000000000..03b951b924be
--- /dev/null
+++ b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_disable-gcc12-warning.patch
@@ -0,0 +1,20 @@
+GCC 12 has a false positive warning when compiled for some architectures, e.g. Intel Sapphire Rapids.
+See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112370
+
+Suppress this warning such that the build doesn't error.
+
+Author: Alexander Grund (TU Dresden)
+
+Ported to 2.3.0, Åke Sandgren
+diff --git a/CMakeLists.txt b/CMakeLists.txt
+index b74bf4536f4..bb062fa843a 100644
+--- a/CMakeLists.txt
++++ b/CMakeLists.txt
+@@ -573,6 +573,7 @@ if(MSVC)
+   string(APPEND CMAKE_CXX_FLAGS " /FS")
+   string(APPEND CMAKE_CUDA_FLAGS " -Xcompiler /FS")
+ endif(MSVC)
++append_cxx_flag_if_supported("-Wno-free-nonheap-object" CMAKE_CXX_FLAGS)
+ 
+ string(APPEND CMAKE_CUDA_FLAGS " -Xfatbin -compress-all")
+ 
diff --git a/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_disable_DataType_dependent_test_if_tensorboard_is_not_available.patch b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_disable_DataType_dependent_test_if_tensorboard_is_not_available.patch
new file mode 100644
index 000000000000..b08d67aaeb49
--- /dev/null
+++ b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_disable_DataType_dependent_test_if_tensorboard_is_not_available.patch
@@ -0,0 +1,126 @@
+Disable use of DataType in test when tensorboard module is not available
+The corresponding import is already protected
+
+Åke Sandgren, 2024-05-07
+diff --git a/test/test_tensorboard.py b/test/test_tensorboard.py
+index 3ce2ab2a172..141f09dc220 100644
+--- a/test/test_tensorboard.py
++++ b/test/test_tensorboard.py
+@@ -829,64 +829,65 @@ class TestTensorBoardNumpy(BaseTestCase):
+         )
+         compare_proto(graph, self)
+ 
+-class TestTensorProtoSummary(BaseTestCase):
+-    @parametrize(
+-        "tensor_type,proto_type",
+-        [
+-            (torch.float16, DataType.DT_HALF),
+-            (torch.bfloat16, DataType.DT_BFLOAT16),
+-        ],
+-    )
+-    def test_half_tensor_proto(self, tensor_type, proto_type):
+-        float_values = [1.0, 2.0, 3.0]
+-        actual_proto = tensor_proto(
+-            "dummy",
+-            torch.tensor(float_values, dtype=tensor_type),
+-        ).value[0].tensor
+-        self.assertSequenceEqual(
+-            [int_to_half(x) for x in actual_proto.half_val],
+-            float_values,
++if TEST_TENSORBOARD:
++    class TestTensorProtoSummary(BaseTestCase):
++        @parametrize(
++            "tensor_type,proto_type",
++            [
++                (torch.float16, DataType.DT_HALF),
++                (torch.bfloat16, DataType.DT_BFLOAT16),
++            ],
+         )
+-        self.assertTrue(actual_proto.dtype == proto_type)
++        def test_half_tensor_proto(self, tensor_type, proto_type):
++            float_values = [1.0, 2.0, 3.0]
++            actual_proto = tensor_proto(
++                "dummy",
++                torch.tensor(float_values, dtype=tensor_type),
++            ).value[0].tensor
++            self.assertSequenceEqual(
++                [int_to_half(x) for x in actual_proto.half_val],
++                float_values,
++            )
++            self.assertTrue(actual_proto.dtype == proto_type)
+ 
+-    def test_float_tensor_proto(self):
+-        float_values = [1.0, 2.0, 3.0]
+-        actual_proto = (
+-            tensor_proto("dummy", torch.tensor(float_values)).value[0].tensor
+-        )
+-        self.assertEqual(actual_proto.float_val, float_values)
+-        self.assertTrue(actual_proto.dtype == DataType.DT_FLOAT)
+-
+-    def test_int_tensor_proto(self):
+-        int_values = [1, 2, 3]
+-        actual_proto = (
+-            tensor_proto("dummy", torch.tensor(int_values, dtype=torch.int32))
+-            .value[0]
+-            .tensor
+-        )
+-        self.assertEqual(actual_proto.int_val, int_values)
+-        self.assertTrue(actual_proto.dtype == DataType.DT_INT32)
++        def test_float_tensor_proto(self):
++            float_values = [1.0, 2.0, 3.0]
++            actual_proto = (
++                tensor_proto("dummy", torch.tensor(float_values)).value[0].tensor
++            )
++            self.assertEqual(actual_proto.float_val, float_values)
++            self.assertTrue(actual_proto.dtype == DataType.DT_FLOAT)
++
++        def test_int_tensor_proto(self):
++            int_values = [1, 2, 3]
++            actual_proto = (
++                tensor_proto("dummy", torch.tensor(int_values, dtype=torch.int32))
++                .value[0]
++                .tensor
++            )
++            self.assertEqual(actual_proto.int_val, int_values)
++            self.assertTrue(actual_proto.dtype == DataType.DT_INT32)
+ 
+-    def test_scalar_tensor_proto(self):
+-        scalar_value = 0.1
+-        actual_proto = (
+-            tensor_proto("dummy", torch.tensor(scalar_value)).value[0].tensor
+-        )
+-        self.assertAlmostEqual(actual_proto.float_val[0], scalar_value)
++        def test_scalar_tensor_proto(self):
++            scalar_value = 0.1
++            actual_proto = (
++                tensor_proto("dummy", torch.tensor(scalar_value)).value[0].tensor
++            )
++            self.assertAlmostEqual(actual_proto.float_val[0], scalar_value)
+ 
+-    def test_complex_tensor_proto(self):
+-        real = torch.tensor([1.0, 2.0])
+-        imag = torch.tensor([3.0, 4.0])
+-        actual_proto = (
+-            tensor_proto("dummy", torch.complex(real, imag)).value[0].tensor
+-        )
+-        self.assertEqual(actual_proto.scomplex_val, [1.0, 3.0, 2.0, 4.0])
++        def test_complex_tensor_proto(self):
++            real = torch.tensor([1.0, 2.0])
++            imag = torch.tensor([3.0, 4.0])
++            actual_proto = (
++                tensor_proto("dummy", torch.complex(real, imag)).value[0].tensor
++            )
++            self.assertEqual(actual_proto.scomplex_val, [1.0, 3.0, 2.0, 4.0])
+ 
+-    def test_empty_tensor_proto(self):
+-        actual_proto = tensor_proto("dummy", torch.empty(0)).value[0].tensor
+-        self.assertEqual(actual_proto.float_val, [])
++        def test_empty_tensor_proto(self):
++            actual_proto = tensor_proto("dummy", torch.empty(0)).value[0].tensor
++            self.assertEqual(actual_proto.float_val, [])
+ 
+-instantiate_parametrized_tests(TestTensorProtoSummary)
++    instantiate_parametrized_tests(TestTensorProtoSummary)
+ 
+ if __name__ == '__main__':
+     run_tests()
diff --git a/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_disable_test_linear_package_if_no_half_types_are_available.patch b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_disable_test_linear_package_if_no_half_types_are_available.patch
new file mode 100644
index 000000000000..f6b08659c647
--- /dev/null
+++ b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_disable_test_linear_package_if_no_half_types_are_available.patch
@@ -0,0 +1,15 @@
+Disable test_linear_package if neither bf16 nor fp16 are available on the CPU
+
+Åke Sandgren, 2024-05-06
+diff --git a/test/inductor/test_cpu_repro.py b/test/inductor/test_cpu_repro.py
+index dfc453220a9..ae1613ad62e 100644
+--- a/test/inductor/test_cpu_repro.py
++++ b/test/inductor/test_cpu_repro.py
+@@ -362,6 +362,7 @@ class CPUReproTests(TestCase):
+             self.common(Model(), example_inputs)
+ 
+     @unittest.skipIf(not torch.backends.mkldnn.is_available(), "MKLDNN is not enabled")
++    @unittest.skipIf(not (torch.ops.mkldnn._is_mkldnn_bf16_supported() or torch.ops.mkldnn._is_mkldnn_fp16_supported()), "MKLDNN neither fp16 nor bf16 are available")
+     @patch("torch.cuda.is_available", lambda: False)
+     def test_linear_packed(self):
+         dtypes = []
diff --git a/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_disable_tests_which_need_network_download.patch b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_disable_tests_which_need_network_download.patch
new file mode 100644
index 000000000000..550733683825
--- /dev/null
+++ b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_disable_tests_which_need_network_download.patch
@@ -0,0 +1,27 @@
+Disable tests that requires external downloads
+
+Åke Sandgren, 2024-05-14
+diff --git a/test/test_hub.py b/test/test_hub.py
+index a0c0c315c4e..6427d02045f 100644
+--- a/test/test_hub.py
++++ b/test/test_hub.py
+@@ -23,6 +23,7 @@ TORCHHUB_EXAMPLE_RELEASE_URL = 'https://github.com/ailzhang/torchhub_example/rel
+ 
+ 
+ @unittest.skipIf(IS_SANDCASTLE, 'Sandcastle cannot ping external')
++@unittest.skipIf(True, 'EasyBuild: do not want tests to depend on downloading')
+ class TestHub(TestCase):
+ 
+     def setUp(self):
+diff --git a/test/test_nn.py b/test/test_nn.py
+index 66fc4cdc466..ed5d1ecc765 100644
+--- a/test/test_nn.py
++++ b/test/test_nn.py
+@@ -151,6 +151,7 @@ class TestNN(NNTestCase):
+             for b in m.buffers():
+                 self.assertFalse(b.requires_grad)
+ 
++    @unittest.skipIf(True, 'EasyBuild: do not want tests to depend on downloading')
+     def test_module_backcompat(self):
+         from torch.serialization import SourceChangeWarning
+         path = download_file('https://download.pytorch.org/test_data/linear.pt')
diff --git a/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_fix-mkldnn-avx512-f32-bias.patch b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_fix-mkldnn-avx512-f32-bias.patch
new file mode 100644
index 000000000000..312c900be477
--- /dev/null
+++ b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_fix-mkldnn-avx512-f32-bias.patch
@@ -0,0 +1,19 @@
+Fix a bug in MKLDNN that leads to e.g. a failure in
+test_conv_deconv_1d_lower_precision_cpu_bfloat16, test_conv_deconv_2d_lower_precision_cpu_bfloat16, test_conv_deconv_3d_lower_precision_cpu_bfloat16
+
+From: https://github.com/uxlfoundation/oneDNN/commit/c145daecbfb285a876a689e665ef0f2875638a60
+From: Dmitrii Zarukin <dmitry.zarukin@intel.com>
+Subject: [PATCH] cpu: x64: avx512_core_amx: kernel: fix f32 bias application
+
+diff -ur a/third_party/ideep/mkl-dnn/src/cpu/x64/jit_avx512_core_amx_conv_kernel.cpp b/third_party/ideep/mkl-dnn/src/cpu/x64/jit_avx512_core_amx_conv_kernel.cpp
+--- a/third_party/ideep/mkl-dnn/src/cpu/x64/jit_avx512_core_amx_conv_kernel.cpp
++++ b/third_party/ideep/mkl-dnn/src/cpu/x64/jit_avx512_core_amx_conv_kernel.cpp
+@@ -3205,7 +3205,7 @@
+                 vpslld(zmm_in, zmm_in, 16);
+                 break;
+             case data_type::f16: vcvtph2ps(zmm_in_k, addr); break;
+-            case data_type::f32: vaddps(zmm_in_k, addr); return;
++            case data_type::f32: vmovups(zmm_in_k, addr); break;
+             default: assert(!"Unsupported data type in xf16 conv");
+         }
+         vaddps(zmm_out, zmm_in);
diff --git a/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_fix-test_extension_backend-without-vectorization.patch b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_fix-test_extension_backend-without-vectorization.patch
new file mode 100644
index 000000000000..c5d78f181d4c
--- /dev/null
+++ b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_fix-test_extension_backend-without-vectorization.patch
@@ -0,0 +1,40 @@
+The test checks for a substring "loadu" in generated code.
+On AVX systems that line is:
+> auto tmp0 = at::vec::Vectorized<float>::loadu(in_ptr0 + static_cast<long>(i0))
+however on non-AVX systems it is
+> auto tmp0 = in_ptr0[static_cast<long>(i0)];
+
+the difference depends on `codecache.valid_vec_isa_list()` being non-empty.
+See torch/_inductor/codegen/cpp.py:2639
+
+Modify the test to account for that.
+
+Author: Alexander Grund (TU Dresden)
+
+Ported to 2.3.0, Åke Sandgren
+diff --git a/test/inductor/test_extension_backend.py b/test/inductor/test_extension_backend.py
+index 54f020907a5..a4b1ce88d99 100644
+--- a/test/inductor/test_extension_backend.py
++++ b/test/inductor/test_extension_backend.py
+@@ -20,7 +20,7 @@ except ImportError:
+     )
+ 
+ from torch._C import FileCheck
+-from torch._inductor import metrics
++from torch._inductor import codecache, metrics
+ from torch._inductor.codegen.common import (
+     get_scheduling_for_device,
+     get_wrapper_codegen_for_device,
+@@ -131,7 +131,11 @@ class ExtensionBackendTests(TestCase):
+         metrics.reset()
+         opt_fn = torch.compile()(fn)
+         _, code = run_and_get_cpp_code(opt_fn, x, y, z)
+-        FileCheck().check("void kernel").check("loadu").check("extension_device").run(
++        if codecache.valid_vec_isa_list():
++            load_expr = 'loadu'
++        else:
++            load_expr = ' = in_ptr0[static_cast<long>(i0)];'
++        FileCheck().check("void kernel").check(load_expr).check("extension_device").run(
+             code
+         )
+         opt_fn(x, y, z)
diff --git a/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_fix_missing_masked_load_for_int_type.patch b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_fix_missing_masked_load_for_int_type.patch
new file mode 100644
index 000000000000..790436918fa9
--- /dev/null
+++ b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_fix_missing_masked_load_for_int_type.patch
@@ -0,0 +1,122 @@
+commit eda279c99771de314d8b82aa8389c9e972f9b855
+Author: Nikita Shulga <nikita.shulga@gmail.com>
+Date:   Mon Mar 25 12:33:59 2024 -0700
+
+    [CpuInductor] Implement masked_load for integral types (#122608)
+    
+    Use `if constexpr` to separate float vs integral masked load for avx512
+    Discovered while looking at `test_comprehensive_fft_ihfft2_cpu_int64` on
+    non-AVX512 capable CPUs where (5, 6, 7) shape were big enough to start a vectorized loop
+    
+    Added `test_pad_cast` regression test
+    
+    Fixes https://github.com/pytorch/pytorch/issues/122606
+    
+    Pull Request resolved: https://github.com/pytorch/pytorch/pull/122608
+    Approved by: https://github.com/jansel
+    ghstack dependencies: #122607
+
+diff --git a/test/inductor/test_torchinductor.py b/test/inductor/test_torchinductor.py
+index 303cc44dd7a..552763e84ae 100644
+--- a/test/inductor/test_torchinductor.py
++++ b/test/inductor/test_torchinductor.py
+@@ -8497,6 +8497,13 @@ class CommonTemplate:
+         x = torch.rand(48, 3, 512, 512)
+         self.common(fn, (x,))
+ 
++    def test_pad_cast(self):
++        def fn(x):
++            return torch.nn.functional.pad(x.to(torch.float32), (0, 3, 0, 0))
++
++        for dtype in [torch.int32, torch.int64]:
++            self.common(fn, (torch.ones(1, 1, 13, dtype=dtype),))
++
+     @unittest.skipIf(not HAS_CPU or not RUN_CPU, "requires C++ compiler")
+     def test_data_type_propogation(self):
+         from torch._dynamo.utils import detect_fake_mode
+diff --git a/torch/_inductor/codegen/cpp_prefix.h b/torch/_inductor/codegen/cpp_prefix.h
+index 9bec9d5a90e..c922274d655 100644
+--- a/torch/_inductor/codegen/cpp_prefix.h
++++ b/torch/_inductor/codegen/cpp_prefix.h
+@@ -280,19 +280,36 @@ inline float flag_to_float_scalar(T src) {
+ 
+ #if defined(CPU_CAPABILITY_AVX512) || defined(CPU_CAPABILITY_AVX2) || defined(CPU_CAPABILITY_ZVECTOR)
+ 
+-inline at::vec::Vectorized<float> masked_load(const float* src, at::vec::Vectorized<float> mask) {
++template<typename T>
++typename std::enable_if_t<std::is_same_v<T, float> ||  std::is_same_v<T, int32_t> || std::is_same_v<T, uint32_t>, at::vec::Vectorized<T>>
++inline masked_load(const T* src, at::vec::Vectorized<float> mask) {
+ # if defined(CPU_CAPABILITY_AVX512)
+-    at::vec::Vectorized<float> zero_vec(0);
++    at::vec::Vectorized<T> zero_vec(0);
+     auto all_ones = _mm512_set1_epi32(0xFFFFFFFF);
+     auto mmask = _mm512_cmp_epi32_mask(_mm512_castps_si512(mask), all_ones, _MM_CMPINT_EQ);
+-    return _mm512_mask_loadu_ps(zero_vec, mmask, src);
++
++    if constexpr (std::is_same_v<T, float>) {
++        return _mm512_mask_loadu_ps(zero_vec, mmask, src);
++    } else {
++        return _mm512_mask_loadu_epi32(zero_vec, mmask, src);
++    }
+ # elif defined(CPU_CAPABILITY_AVX2)
+     auto all_ones = _mm256_set1_epi32(0xFFFFFFFF);
+     auto mmask = _mm256_cmpeq_epi32(_mm256_castps_si256(mask), all_ones);
+-    return _mm256_maskload_ps(src, mmask);
++    if constexpr (std::is_same_v<T, float>) {
++        return _mm256_maskload_ps(src, mmask);
++    } else {
++        return _mm256_maskload_epi32(src, mmask);
++    }
+ # elif defined(CPU_CAPABILITY_ZVECTOR)
+-    auto result = at::vec::Vectorized<float>::loadu(src);
+-    return (result & mask);
++    auto result = at::vec::Vectorized<T>::loadu(src);
++    if constexpr (std::is_same_v<T, float>) {
++        return result & mask;
++    } else {
++        T maskdata[at::vec::Vectorized<T>::size()];
++        mask.store(maskdata);
++        return result & at::vec::Vectorized<T>::loadu(maskdata);
++    }
+ # else
+ # error Unsupported vectorization CPU capability
+ # endif
+@@ -326,7 +343,7 @@ inline masked_load(const T* src, at::vec::Vectorized<float> mask) {
+     maskdata_dest[i] = (maskdata[i] == 0xFFFFFFFF) ? 0xFFFF: 0;
+   }
+   auto maskvector = at::vec::Vectorized<T>::loadu(maskdata_dest);
+-  return (result & maskvector);
++  return result & maskvector;
+ # else
+ # error Unsupported vectorization CPU capability
+ # endif
+@@ -360,12 +377,28 @@ inline masked_load(const T* src, at::vec::Vectorized<float> mask) {
+       maskdata_dest[i] = (maskdata[i] == 0xFFFFFFFF) ? 0xFF: 0;
+     }
+     auto maskvector = at::vec::Vectorized<T>::loadu(maskdata_dest);
+-    return (result & maskvector);
++    return result & maskvector;
+ # else
+ # error Unsupported vectorization CPU capability
+ # endif
+ }
+ 
++template <typename T>
++typename std::enable_if_t<std::is_same_v<T, uint64_t> || std::is_same_v<T, int64_t>, at::vec::VectorizedN<T, 2>>
++inline masked_load(const T* src, at::vec::Vectorized<float> mask) {
++  // TODO: Add vectorized variants for the load
++  constexpr auto vec_size = decltype(mask)::size();
++  __at_align__ uint32_t maskdata[vec_size];
++  __at_align__ uint64_t mask_res[vec_size];
++  mask.store(maskdata);
++  #pragma unroll
++  for(auto i = 0; i < vec_size; ++i) {
++          mask_res[i] = maskdata[i] ? std::numeric_limits<uint64_t>::max() : 0;
++  }
++  auto result = at::vec::VectorizedN<T, 2>::loadu(src);
++  return result & at::vec::VectorizedN<T, 2>::loadu(mask_res);
++}
++
+ template <typename T>
+ inline at::vec::Vectorized<float> flag_to_float_vec(const T* src) {
+   __at_align__ float dst_tmp[at::vec::Vectorized<float>::size()];
diff --git a/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_no-cuda-stubs-rpath.patch b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_no-cuda-stubs-rpath.patch
new file mode 100644
index 000000000000..20d6ed8a80dc
--- /dev/null
+++ b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_no-cuda-stubs-rpath.patch
@@ -0,0 +1,150 @@
+# PyTorch's CMAKE configuration by default sets RUNPATH on libraries if they link other libraries
+# that are outside the build tree, which is done because of the CMAKE config on 
+# https://github.com/pytorch/pytorch/blob/v1.10.0/cmake/Dependencies.cmake#L10.
+# This provides problems, since the cuda stubs library path then also gets added to the RUNPATH.
+# As a result, at runtime, the stub version of things like libcuda.so.1 gets picked up, instead of the real drivers
+# See https://github.com/easybuilders/easybuild-easyconfigs/issues/14359
+# This line https://github.com/pytorch/pytorch/blob/v1.10.0/cmake/Dependencies.cmake#L16
+# Makes sure that any path that is linked, is also added to the RUNPATH.
+# This has been reported upstream in https://github.com/pytorch/pytorch/issues/35418
+# and a fix was attempted in https://github.com/pytorch/pytorch/pull/37737 but it was reverted
+#
+# This EasyBuild patch changes behavior for the libraries that were failing, i.e. the ones in this list:
+# https://github.com/easybuilders/easybuild-easyconfigs/issues/14359#issuecomment-970479904
+# This is done by setting INSTALL_RPATH_USE_LINK_PATH to false, and instead, specifying the RPATH
+# explicitely by defining INSTALL_RPATH, but only adding directories that do not match to the "stubs" regex
+#
+# Original patch: Caspar van Leeuwen
+# Updated: Alexander Grund (TU Dresden)
+#
+# See https://github.com/pytorch/pytorch/pull/122318
+
+diff --git a/binaries/CMakeLists.txt b/binaries/CMakeLists.txt
+index 15f47bf52ae..edf1ab26149 100644
+--- a/binaries/CMakeLists.txt
++++ b/binaries/CMakeLists.txt
+@@ -56,7 +56,8 @@ endif()
+ 
+ if(USE_CUDA)
+   caffe2_binary_target("inspect_gpu.cc")
+-  target_link_libraries(inspect_gpu ${CUDA_LIBRARIES})
++  include(${Torch_SOURCE_DIR}/cmake/LinkCudaLibraries.cmake)
++  link_cuda_libraries(inspect_gpu ${CUDA_LIBRARIES})
+   caffe2_binary_target("print_core_object_sizes_gpu.cc")
+ 
+   if(BUILD_TEST)
+diff --git a/caffe2/CMakeLists.txt b/caffe2/CMakeLists.txt
+index 5a68e225155..bb87d973d1a 100644
+--- a/caffe2/CMakeLists.txt
++++ b/caffe2/CMakeLists.txt
+@@ -651,14 +651,13 @@ endif()
+ if(USE_CUDA)
+   list(APPEND Caffe2_GPU_CU_SRCS ${Caffe2_GPU_HIP_JIT_FUSERS_SRCS})
+   add_library(caffe2_nvrtc SHARED ${ATen_NVRTC_STUB_SRCS})
++  include(${Torch_SOURCE_DIR}/cmake/LinkCudaLibraries.cmake)
++  link_cuda_libraries(caffe2_nvrtc ${CUDA_CUDA_LIB} ${CUDA_NVRTC_LIB})
+   if(MSVC)
+     # Delay load nvcuda.dll so we can import torch compiled with cuda on a CPU-only machine
+-    set(DELAY_LOAD_FLAGS "-DELAYLOAD:nvcuda.dll;delayimp.lib")
+-  else()
+-    set(DELAY_LOAD_FLAGS "")
++    target_link_libraries(caffe2_nvrtc "-DELAYLOAD:nvcuda.dll;delayimp.lib")
+   endif()
+ 
+-  target_link_libraries(caffe2_nvrtc ${CUDA_CUDA_LIB} ${CUDA_NVRTC_LIB} ${DELAY_LOAD_FLAGS})
+   install(TARGETS caffe2_nvrtc DESTINATION "${TORCH_INSTALL_LIB_DIR}")
+   if(USE_NCCL)
+     list(APPEND Caffe2_GPU_SRCS
+@@ -1607,6 +1606,7 @@ endif()
+ 
+ # ---[ CUDA library.
+ if(USE_CUDA)
++  include(${Torch_SOURCE_DIR}/cmake/LinkCudaLibraries.cmake)
+   # FIXME: If kineto is linked with CUPTI it pollutes torch_cpu with CUDA dependencies
+   # Even worse, it never declares that it depends on cudart, but calls the API, see
+   # https://github.com/pytorch/kineto/blob/aef2f5c0f15e3be52406ac0b885e8689de6bc9f6/libkineto/src/CudaDeviceProperties.cpp#L24
+@@ -1620,13 +1620,13 @@ if(USE_CUDA)
+       torch_cuda INTERFACE $<INSTALL_INTERFACE:include>)
+   target_include_directories(
+       torch_cuda PRIVATE ${Caffe2_GPU_INCLUDE})
+-  target_link_libraries(
++  link_cuda_libraries(
+       torch_cuda PRIVATE ${Caffe2_CUDA_DEPENDENCY_LIBS})
+ 
+   # These public dependencies must go after the previous dependencies, as the
+   # order of the libraries in the linker call matters here when statically
+   # linking; libculibos and cublas must be last.
+-  target_link_libraries(torch_cuda PUBLIC torch_cpu_library ${Caffe2_PUBLIC_CUDA_DEPENDENCY_LIBS})
++  link_cuda_libraries(torch_cuda PUBLIC torch_cpu_library ${Caffe2_PUBLIC_CUDA_DEPENDENCY_LIBS})
+ endif()
+ 
+ # ---[ XPU library.
+diff --git a/cmake/LinkCudaLibraries.cmake b/cmake/LinkCudaLibraries.cmake
+new file mode 100644
+index 00000000000..e09d1186f6d
+--- /dev/null
++++ b/cmake/LinkCudaLibraries.cmake
+@@ -0,0 +1,33 @@
++# Link CUDA libraries to the given target, i.e.: `target_link_libraries(target <args>)`
++#
++# Additionally makes sure CUDA stub libs don't end up being in RPath
++#
++# Example: link_cuda_libraries(mytarget PRIVATE ${CUDA_LIBRARIES})
++function(link_cuda_libraries target)
++  set(libs ${ARGN})
++  set(install_rpath "$ORIGIN")
++  set(filtered FALSE)
++  foreach(lib IN LISTS libs)
++    # CUDA stub libs are in form /prefix/lib/stubs/libcuda.so
++    # So extract the name of the parent folder, to check against "stubs"
++    # And the parent path which we need to add to the INSTALL_RPATH for non-stubs
++    get_filename_component(parent_path "${lib}" DIRECTORY)
++    get_filename_component(parent_name "${parent_path}" NAME)
++    if(parent_name STREQUAL "stubs")
++      message(STATUS "Filtering ${lib} from being set in ${target}'s RPATH, "
++                     "because it appears to point to the CUDA stubs directory.")
++      set(filtered TRUE)
++    elseif(parent_path)
++      list(APPEND install_rpath ${parent_path})
++    endif()
++  endforeach()
++
++  # Regular link command
++  target_link_libraries(${target} ${libs})
++  # Manually set INSTALL_RPATH when there were any stub libs
++  if(filtered)
++    list(REMOVE_DUPLICATES install_rpath)
++    set_target_properties(${target} PROPERTIES INSTALL_RPATH_USE_LINK_PATH FALSE)
++    set_target_properties(${target} PROPERTIES INSTALL_RPATH "${install_rpath}")
++  endif()
++endfunction()
+diff --git a/test/test_torch.py b/test/test_torch.py
+index 9a1b6199032..f15dbc7b9a4 100644
+--- a/test/test_torch.py
++++ b/test/test_torch.py
+@@ -10730,6 +10730,25 @@ def add_neg_dim_tests():
+         assert not hasattr(TestTorch, test_name), "Duplicated test name: " + test_name
+         setattr(TestTorch, test_name, make_neg_dim_test(name, tensor_arg, arg_constr, types, extra_dim))
+ 
++class TestRPATH(TestCase):
++    @unittest.skipIf(not sys.platform.startswith('linux'), "linux-only test")
++    def test_rpath(self):
++        """
++        Make sure RPATH (or RUNPATH) in nvrtc does not contain a cuda stubs directory
++        issue gh-35418
++        """
++        libdir = os.path.join(os.path.dirname(torch._C.__file__), 'lib')
++        caffe2_nvrtc = os.path.join(libdir, 'libcaffe2_nvrtc.so')
++        if os.path.exists(caffe2_nvrtc):
++            output = subprocess.check_output(['objdump', '-x', caffe2_nvrtc])
++            for line in output.split(b'\n'):
++                if b'RPATH' in line or b'RUNPATH' in line:
++                    # Catch paths like:
++                    # $ORIGIN:/usr/local/cuda-10.2/lib64/stubs:/usr/local/cuda-10.2/lib64
++                    self.assertFalse(b'/stubs:' in line)
++                    # $ORIGIN:/usr/local/cuda-10.2/lib64:/usr/local/cuda-10.2/lib64/stubs
++                    self.assertFalse(line.endswith(b'/stubs'))
++
+ # TODO: these empy classes are temporarily instantiated for XLA compatibility
+ #   once XLA updates their test suite it should be removed
+ class TestViewOps(TestCase):
diff --git a/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_skip-test_init_from_local_shards.patch b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_skip-test_init_from_local_shards.patch
new file mode 100644
index 000000000000..6da0ec670202
--- /dev/null
+++ b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_skip-test_init_from_local_shards.patch
@@ -0,0 +1,26 @@
+The test often times out and seems to be considered flaky by PyTorch:
+https://github.com/pytorch/pytorch/issues/78068
+
+Author: Alexander Grund (TU Dresden)
+
+Ported to 2.3.0, Åke Sandgren
+diff --git a/test/distributed/_shard/sharded_tensor/test_sharded_tensor.py b/test/distributed/_shard/sharded_tensor/test_sharded_tensor.py
+index f2abc1e92d6..f32a298e813 100644
+--- a/test/distributed/_shard/sharded_tensor/test_sharded_tensor.py
++++ b/test/distributed/_shard/sharded_tensor/test_sharded_tensor.py
+@@ -7,6 +7,7 @@ import itertools
+ import pickle
+ import sys
+ from typing import List
++from unittest import skip
+ import torch
+ import torch.distributed as dist
+ from torch.distributed import rpc
+@@ -2267,6 +2268,7 @@ class TestShardedTensorFromLocalShards(ShardedTensorTestBase):
+     @with_comms
+     @skip_if_lt_x_gpu(4)
+     @requires_nccl()
++    @skip("Times out often")
+     def test_init_from_local_shards(self):
+         local_shard_metadata = ShardMetadata(
+             shard_offsets=[(self.rank // 2) * 5, (self.rank % 2) * 5],
diff --git a/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_skip_test_sdpa_nn_functional_scaled_dot_product_attention_cpu.patch b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_skip_test_sdpa_nn_functional_scaled_dot_product_attention_cpu.patch
new file mode 100644
index 000000000000..a1e29ddd8715
--- /dev/null
+++ b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_skip_test_sdpa_nn_functional_scaled_dot_product_attention_cpu.patch
@@ -0,0 +1,20 @@
+commit 8839fc56214ff6154ae181fb8122989752f270b3
+Author: Ake Sandgren <ake.sandgren@hpc2n.umu.se>
+Date:   Thu Mar 13 07:22:52 2025 +0100
+
+    Skip test_sdpa_nn_functional_scaled_dot_product_attention_cpu_half/bfloat16
+    It fails consistently on Rapids and Rome
+
+diff --git a/test/test_decomp.py b/test/test_decomp.py
+index 4361895e15a..d18c201c360 100644
+--- a/test/test_decomp.py
++++ b/test/test_decomp.py
+@@ -932,7 +932,7 @@ class DecompOneOffTests(TestCase):
+     @onlyCPU
+     @skipIfCrossRef
+     @skipOps('DecompOneOffTests', 'test_sdpa', [
+-        xfail("nn.functional.scaled_dot_product_attention", dtypes=[torch.half] + ([torch.bfloat16] if IS_MACOS else [])),
++        skip("nn.functional.scaled_dot_product_attention", dtypes=[torch.half, torch.bfloat16]),
+     ])
+     @ops(_sdpa_op_info)
+     def test_sdpa(self, device, dtype, op):
diff --git a/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_skip_test_var_mean_differentiable.patch b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_skip_test_var_mean_differentiable.patch
new file mode 100644
index 000000000000..c7c2c408be06
--- /dev/null
+++ b/easybuild/easyconfigs/p/PyTorch/PyTorch-2.3.0_skip_test_var_mean_differentiable.patch
@@ -0,0 +1,19 @@
+commit 46010b9d50aed454e9c9e5347df10e6cbb6eac39
+Author: Ake Sandgren <ake.sandgren@hpc2n.umu.se>
+Date:   Mon Mar 10 11:13:33 2025 +0100
+
+    Skip test_var_mean_differentiable
+    It sometimoes gets a segfault during garbage collection causing problems in PASS/FAIL reporting
+
+diff --git a/test/test_autograd.py b/test/test_autograd.py
+index 791510e4cc4..94a826c11ec 100644
+--- a/test/test_autograd.py
++++ b/test/test_autograd.py
+@@ -3881,6 +3881,7 @@ class TestAutograd(TestCase):
+         with self.assertRaisesRegex(Exception, 'Simulate error'):
+             d.sum().backward()
+ 
++    @unittest.skip("Occasionally gets a segfault during garbage collection")
+     def test_var_mean_differentiable(self):
+         dim = [2, 4]
+         keepdim = False
diff --git a/easybuild/easyconfigs/t/tlparse/tlparse-0.3.5-GCCcore-13.2.0.eb b/easybuild/easyconfigs/t/tlparse/tlparse-0.3.5-GCCcore-13.2.0.eb
new file mode 100644
index 000000000000..c233755fa537
--- /dev/null
+++ b/easybuild/easyconfigs/t/tlparse/tlparse-0.3.5-GCCcore-13.2.0.eb
@@ -0,0 +1,185 @@
+easyblock = 'CargoPythonPackage'
+
+name = 'tlparse'
+version = '0.3.5'
+
+homepage = 'https://github.com/ezyang/tlparse'
+description = "Parse structured PT2 (PyTorch 2) logs"
+
+toolchain = {'name': 'GCCcore', 'version': '13.2.0'}
+
+github_account = 'ezyang'
+source_urls = [GITHUB_SOURCE]
+sources = ['v%(version)s.tar.gz']
+checksums = [
+    {'v0.3.5.tar.gz': '256eb651bc435c75ff33ad4a31695db99e7e8441d45d2a3124c66fbe2a4b1d30'},
+    {'aho-corasick-1.1.2.tar.gz': 'b2969dcb958b36655471fc61f7e416fa76033bdd4bfed0678d8fee1e2d07a1f0'},
+    {'anstream-0.6.13.tar.gz': 'd96bd03f33fe50a863e394ee9718a706f988b9079b20c3784fb726e7678b62fb'},
+    {'anstyle-1.0.6.tar.gz': '8901269c6307e8d93993578286ac0edf7f195079ffff5ebdeea6a59ffb7e36bc'},
+    {'anstyle-parse-0.2.3.tar.gz': 'c75ac65da39e5fe5ab759307499ddad880d724eed2f6ce5b5e8a26f4f387928c'},
+    {'anstyle-query-1.0.2.tar.gz': 'e28923312444cdd728e4738b3f9c9cac739500909bb3d3c94b43551b16517648'},
+    {'anstyle-wincon-3.0.2.tar.gz': '1cd54b81ec8d6180e24654d0b371ad22fc3dd083b6ff8ba325b72e00c87660a7'},
+    {'base16ct-0.2.0.tar.gz': '4c7f02d4ea65f2c1853089ffd8d2787bdbc63de2f0d29dedbcf8ccdfa0ccd4cf'},
+    {'block-buffer-0.10.4.tar.gz': '3078c7629b62d3f0439517fa394996acacc5cbc91c5a20d8c658e77abd503a71'},
+    {'bstr-1.9.1.tar.gz': '05efc5cfd9110c8416e471df0e96702d58690178e206e61b7173706673c93706'},
+    {'byteorder-1.5.0.tar.gz': '1fd0f2584146f6f2ef48085050886acf353beff7305ebd1ae69500e27c67f64b'},
+    {'cfg-if-1.0.0.tar.gz': 'baf1de4339761588bc0619e3cbc0120ee582ebb74b53b4efbf79117bd2da40fd'},
+    {'clap-4.5.1.tar.gz': 'c918d541ef2913577a0f9566e9ce27cb35b6df072075769e0b26cb5a554520da'},
+    {'clap_builder-4.5.1.tar.gz': '9f3e7391dad68afb0c2ede1bf619f579a3dc9c2ec67f089baa397123a2f3d1eb'},
+    {'clap_derive-4.5.0.tar.gz': '307bc0538d5f0f83b8248db3087aa92fe504e4691294d0c96c0eabc33f47ba47'},
+    {'clap_lex-0.7.0.tar.gz': '98cc8fbded0c607b7ba9dd60cd98df59af97e84d24e49c8557331cfc26d301ce'},
+    {'colorchoice-1.0.0.tar.gz': 'acbf1af155f9b9ef647e42cdc158db4b64a1b61f743629225fde6f3e0be2a7c7'},
+    {'console-0.15.8.tar.gz': '0e1f83fc076bd6dd27517eacdf25fef6c4dfe5f1d7448bafaaf3a26f13b5e4eb'},
+    {'crypto-common-0.1.6.tar.gz': '1bfb12502f3fc46cca1bb51ac28df9d618d813cdc3d2f25b9fe775a34af26bb3'},
+    {'digest-0.10.7.tar.gz': '9ed9a281f7bc9b7576e61468ba615a66a5c8cfdff42420a70aa82701a3b1e292'},
+    {'encode_unicode-0.3.6.tar.gz': 'a357d28ed41a50f9c765dbfe56cbc04a64e53e5fc58ba79fbc34c10ef3df831f'},
+    {'equivalent-1.0.1.tar.gz': '5443807d6dff69373d433ab9ef5378ad8df50ca6298caf15de6e52e24aaf54d5'},
+    {'fxhash-0.2.1.tar.gz': 'c31b6d751ae2c7f11320402d34e41349dd1016f8d5d45e48c4312bc8625af50c'},
+    {'generic-array-0.14.7.tar.gz': '85649ca51fd72272d7821adaf274ad91c288277713d9c18820d8499a7ff69e9a'},
+    {'hashbrown-0.14.3.tar.gz': '290f1a1d9242c78d09ce40a5e87e7554ee637af1351968159f4952f028f75604'},
+    {'heck-0.4.1.tar.gz': '95505c38b4572b2d910cecb0281560f54b440a19336cbbcb27bf6ce6adc6f5a8'},
+    {'html-escape-0.2.13.tar.gz': '6d1ad449764d627e22bfd7cd5e8868264fc9236e07c752972b4080cd351cb476'},
+    {'indexmap-2.2.3.tar.gz': '233cf39063f058ea2caae4091bf4a3ef70a653afbc026f5c4a4135d114e3c177'},
+    {'indicatif-0.17.8.tar.gz': '763a5a8f45087d6bcea4222e7b72c291a054edf80e4ef6efd2a4979878c7bea3'},
+    {'instant-0.1.12.tar.gz': '7a5bbe824c507c5da5956355e86a746d82e0e1464f65d862cc5e71da70e94b2c'},
+    {'itoa-1.0.10.tar.gz': 'b1a46d1a171d865aa5f83f92695765caa047a9b4cbae2cbf37dbd613a793fd4c'},
+    {'lazy_static-1.4.0.tar.gz': 'e2abad23fbc42b3700f2f279844dc832adb2b2eb069b2df918f455c4e18cc646'},
+    {'libc-0.2.153.tar.gz': '9c198f91728a82281a64e1f4f9eeb25d82cb32a5de251c6bd1b5154d63a8e7bd'},
+    {'md-5-0.10.6.tar.gz': 'd89e7ee0cfbedfc4da3340218492196241d89eefb6dab27de5df917a6d2e78cf'},
+    {'memchr-2.7.1.tar.gz': '523dc4f511e55ab87b694dc30d0f820d60906ef06413f93d4d7a1385599cc149'},
+    {'normpath-1.2.0.tar.gz': '5831952a9476f2fed74b77d74182fa5ddc4d21c72ec45a333b250e3ed0272804'},
+    {'number_prefix-0.4.0.tar.gz': '830b246a0e5f20af87141b25c173cd1b609bd7779a4617d6ec582abaf90870f3'},
+    {'once_cell-1.19.0.tar.gz': '3fdb12b2476b595f9358c5161aa467c2438859caa136dec86c26fdd2efe17b92'},
+    {'opener-0.6.1.tar.gz': '6c62dcb6174f9cb326eac248f07e955d5d559c272730b6c03e396b443b562788'},
+    {'portable-atomic-1.6.0.tar.gz': '7170ef9988bc169ba16dd36a7fa041e5c4cbeb6a35b76d4c03daded371eae7c0'},
+    {'proc-macro2-1.0.78.tar.gz': 'e2422ad645d89c99f8f3e6b88a9fdeca7fabeac836b1002371c4367c8f984aae'},
+    {'quote-1.0.35.tar.gz': '291ec9ab5efd934aaf503a6466c5d5251535d108ee747472c3977cc5acc868ef'},
+    {'regex-1.10.3.tar.gz': 'b62dbe01f0b06f9d8dc7d49e05a0785f153b00b2c227856282f671e0318c9b15'},
+    {'regex-automata-0.4.5.tar.gz': '5bb987efffd3c6d0d8f5f89510bb458559eab11e4f869acb20bf845e016259cd'},
+    {'regex-syntax-0.8.2.tar.gz': 'c08c74e62047bb2de4ff487b251e4a92e24f48745648451635cec7d591162d9f'},
+    {'ryu-1.0.17.tar.gz': 'e86697c916019a8588c99b5fac3cead74ec0b4b819707a682fd4d23fa0ce1ba1'},
+    {'serde-1.0.197.tar.gz': '3fb1c873e1b9b056a4dc4c0c198b24c3ffa059243875552b2bd0933b1aee4ce2'},
+    {'serde_derive-1.0.197.tar.gz': '7eb0b34b42edc17f6b7cac84a52a1c5f0e1bb2227e997ca9011ea3dd34e8610b'},
+    {'serde_json-1.0.114.tar.gz': 'c5f09b1bd632ef549eaa9f60a1f8de742bdbc698e6cee2095fc84dde5f549ae0'},
+    {'strsim-0.11.0.tar.gz': '5ee073c9e4cd00e28217186dbe12796d692868f432bf2e97ee73bed0c56dfa01'},
+    {'syn-2.0.51.tar.gz': '6ab617d94515e94ae53b8406c628598680aa0c9587474ecbe58188f7b345d66c'},
+    {'tinytemplate-1.2.1.tar.gz': 'be4d6b5f19ff7664e8c98d03e2139cb510db9b0a60b55f8e8709b689d939b6bc'},
+    {'typenum-1.17.0.tar.gz': '42ff0bf0c66b8238c6f3b578df37d0b7848e55df8577b3f74f92a69acceeb825'},
+    {'unicode-ident-1.0.12.tar.gz': '3354b9ac3fae1ff6755cb6db53683adb661634f67557942dea4facebec0fee4b'},
+    {'unicode-width-0.1.11.tar.gz': 'e51733f11c9c4f72aa0c160008246859e340b00807569a0da0e7a1079b27ba85'},
+    {'utf8-width-0.1.7.tar.gz': '86bd8d4e895da8537e5315b8254664e6b769c4ff3db18321b297a1e7004392e3'},
+    {'utf8parse-0.2.1.tar.gz': '711b9620af191e0cdc7468a8d14e709c3dcdb115b36f838e601583af800a370a'},
+    {'version_check-0.9.4.tar.gz': '49874b5167b65d7193b8aba1567f5c7d93d001cafc34600cee003eda787e483f'},
+    {'winapi-0.3.9.tar.gz': '5c839a674fcd7a98952e593242ea400abe93992746761e38641405d28b00f419'},
+    {'winapi-i686-pc-windows-gnu-0.4.0.tar.gz': 'ac3b87c63620426dd9b991e5ce0329eff545bccbbb34f3be09ff6fb6ab51b7b6'},
+    {'winapi-x86_64-pc-windows-gnu-0.4.0.tar.gz': '712e227841d057c1ee1cd2fb22fa7e5a5461ae8e48fa2ca79ec42cfc1931183f'},
+    {'windows-sys-0.52.0.tar.gz': '282be5f36a8ce781fad8c8ae18fa3f9beff57ec1b52cb3de0789201425d9a33d'},
+    {'windows-targets-0.52.3.tar.gz': 'd380ba1dc7187569a8a9e91ed34b8ccfc33123bbacb8c0aed2d1ad7f3ef2dc5f'},
+    {'windows_aarch64_gnullvm-0.52.3.tar.gz': '68e5dcfb9413f53afd9c8f86e56a7b4d86d9a2fa26090ea2dc9e40fba56c6ec6'},
+    {'windows_aarch64_msvc-0.52.3.tar.gz': '8dab469ebbc45798319e69eebf92308e541ce46760b49b18c6b3fe5e8965b30f'},
+    {'windows_i686_gnu-0.52.3.tar.gz': '2a4e9b6a7cac734a8b4138a4e1044eac3404d8326b6c0f939276560687a033fb'},
+    {'windows_i686_msvc-0.52.3.tar.gz': '28b0ec9c422ca95ff34a78755cfa6ad4a51371da2a5ace67500cf7ca5f232c58'},
+    {'windows_x86_64_gnu-0.52.3.tar.gz': '704131571ba93e89d7cd43482277d6632589b18ecf4468f591fbae0a8b101614'},
+    {'windows_x86_64_gnullvm-0.52.3.tar.gz': '42079295511643151e98d61c38c0acc444e52dd42ab456f7ccfd5152e8ecf21c'},
+    {'windows_x86_64_msvc-0.52.3.tar.gz': '0770833d60a970638e989b3fa9fd2bb1aaadcf88963d1659fd7d9990196ed2d6'},
+]
+
+builddependencies = [
+    ('binutils', '2.40'),
+    ('maturin', '1.5.0', '-Rust-1.76.0'),
+]
+
+dependencies = [
+    ('Python', '3.11.5'),
+]
+
+use_pip = True
+sanity_pip_check = True
+download_dep_fail = True
+
+options = {
+    'modulename': False,
+}
+
+crates = [
+    ('aho-corasick', '1.1.2'),
+    ('anstream', '0.6.13'),
+    ('anstyle', '1.0.6'),
+    ('anstyle-parse', '0.2.3'),
+    ('anstyle-query', '1.0.2'),
+    ('anstyle-wincon', '3.0.2'),
+    ('base16ct', '0.2.0'),
+    ('block-buffer', '0.10.4'),
+    ('bstr', '1.9.1'),
+    ('byteorder', '1.5.0'),
+    ('cfg-if', '1.0.0'),
+    ('clap', '4.5.1'),
+    ('clap_builder', '4.5.1'),
+    ('clap_derive', '4.5.0'),
+    ('clap_lex', '0.7.0'),
+    ('colorchoice', '1.0.0'),
+    ('console', '0.15.8'),
+    ('crypto-common', '0.1.6'),
+    ('digest', '0.10.7'),
+    ('encode_unicode', '0.3.6'),
+    ('equivalent', '1.0.1'),
+    ('fxhash', '0.2.1'),
+    ('generic-array', '0.14.7'),
+    ('hashbrown', '0.14.3'),
+    ('heck', '0.4.1'),
+    ('html-escape', '0.2.13'),
+    ('indexmap', '2.2.3'),
+    ('indicatif', '0.17.8'),
+    ('instant', '0.1.12'),
+    ('itoa', '1.0.10'),
+    ('lazy_static', '1.4.0'),
+    ('libc', '0.2.153'),
+    ('md-5', '0.10.6'),
+    ('memchr', '2.7.1'),
+    ('normpath', '1.2.0'),
+    ('number_prefix', '0.4.0'),
+    ('once_cell', '1.19.0'),
+    ('opener', '0.6.1'),
+    ('portable-atomic', '1.6.0'),
+    ('proc-macro2', '1.0.78'),
+    ('quote', '1.0.35'),
+    ('regex', '1.10.3'),
+    ('regex-automata', '0.4.5'),
+    ('regex-syntax', '0.8.2'),
+    ('ryu', '1.0.17'),
+    ('serde', '1.0.197'),
+    ('serde_derive', '1.0.197'),
+    ('serde_json', '1.0.114'),
+    ('strsim', '0.11.0'),
+    ('syn', '2.0.51'),
+    ('tinytemplate', '1.2.1'),
+    ('typenum', '1.17.0'),
+    ('unicode-ident', '1.0.12'),
+    ('unicode-width', '0.1.11'),
+    ('utf8-width', '0.1.7'),
+    ('utf8parse', '0.2.1'),
+    ('version_check', '0.9.4'),
+    ('winapi', '0.3.9'),
+    ('winapi-i686-pc-windows-gnu', '0.4.0'),
+    ('winapi-x86_64-pc-windows-gnu', '0.4.0'),
+    ('windows-sys', '0.52.0'),
+    ('windows-targets', '0.52.3'),
+    ('windows_aarch64_gnullvm', '0.52.3'),
+    ('windows_aarch64_msvc', '0.52.3'),
+    ('windows_i686_gnu', '0.52.3'),
+    ('windows_i686_msvc', '0.52.3'),
+    ('windows_x86_64_gnu', '0.52.3'),
+    ('windows_x86_64_gnullvm', '0.52.3'),
+    ('windows_x86_64_msvc', '0.52.3'),
+]
+
+sanity_check_paths = {
+    'files': ['bin/tlparse'],
+    'dirs': [],
+}
+
+sanity_check_commands = [
+    'tlparse --version',
+]
+
+moduleclass = 'tools'