Fix strided slice_scatter KV cache updates #645
Annotations
8 errors and 14 warnings
|
Run Script
Process completed with exit code 1.
|
|
Run Script:
pytorch/tensorrt/tests/py/dynamo/tests/py/dynamo/lowering/test_constant_fold_exclusions.py#L96
TestConstantFoldExclusionMechanics.test_unknown_disabled_rule_is_rejected
self = <dynamo.lowering.test_constant_fold_exclusions.TestConstantFoldExclusionMechanics testMethod=test_unknown_disabled_rule_is_rejected>
def test_unknown_disabled_rule_is_rejected(self):
with self.assertRaisesRegex(
ValueError,
"Unknown constant-fold exclusion rule IDs",
):
> CompilationSettings(disabled_constant_fold_exclusions={"unknown_rule"})
lowering/test_constant_fold_exclusions.py:101:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
<string>:14: in __init__
???
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch_tensorrt/dynamo/_defaults.py:111: in default_device
return Device(gpu_id=torch.cuda.current_device())
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch/cuda/__init__.py:1267: in current_device
_lazy_init()
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
def _lazy_init():
global _initialized, _queued_calls
if is_initialized() or hasattr(_tls, "is_initializing"):
return
with _initialization_lock:
# We be double-checked locking, boys! This is OK because
# the above test was GIL protected anyway. The inner test
# is for when a thread blocked on some other thread which was
# doing the initialization; when they get the lock, they will
# find there is nothing left to do.
if is_initialized():
return
# It is important to prevent other threads from entering _lazy_init
# immediately, while we are still guaranteed to have the GIL, because some
# of the C calls we make below will release the GIL
if _is_in_bad_fork():
> raise RuntimeError(
"Cannot re-initialize CUDA in forked subprocess. To use CUDA with "
"multiprocessing, you must use the 'spawn' start method"
)
E RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method
E
E To execute this test, run the following from the base repo dir:
E python test_constant_fold_exclusions.py TestConstantFoldExclusionMechanics.test_unknown_disabled_rule_is_rejected
E
E This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch/cuda/__init__.py:579: RuntimeError
|
|
Run Script:
pytorch/tensorrt/tests/py/dynamo/tests/py/dynamo/lowering/test_constant_fold_exclusions.py#L65
TestConstantFoldExclusionMechanics.test_setting_changes_engine_compatibility
self = <dynamo.lowering.test_constant_fold_exclusions.TestConstantFoldExclusionMechanics testMethod=test_setting_changes_engine_compatibility>
def test_setting_changes_engine_compatibility(self):
compatible, incompatible_settings = settings_are_compatible(
> CompilationSettings(),
CompilationSettings(disabled_constant_fold_exclusions={self.rule_id}),
)
lowering/test_constant_fold_exclusions.py:67:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
<string>:14: in __init__
???
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch_tensorrt/dynamo/_defaults.py:111: in default_device
return Device(gpu_id=torch.cuda.current_device())
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch/cuda/__init__.py:1267: in current_device
_lazy_init()
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
def _lazy_init():
global _initialized, _queued_calls
if is_initialized() or hasattr(_tls, "is_initializing"):
return
with _initialization_lock:
# We be double-checked locking, boys! This is OK because
# the above test was GIL protected anyway. The inner test
# is for when a thread blocked on some other thread which was
# doing the initialization; when they get the lock, they will
# find there is nothing left to do.
if is_initialized():
return
# It is important to prevent other threads from entering _lazy_init
# immediately, while we are still guaranteed to have the GIL, because some
# of the C calls we make below will release the GIL
if _is_in_bad_fork():
> raise RuntimeError(
"Cannot re-initialize CUDA in forked subprocess. To use CUDA with "
"multiprocessing, you must use the 'spawn' start method"
)
E RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method
E
E To execute this test, run the following from the base repo dir:
E python test_constant_fold_exclusions.py TestConstantFoldExclusionMechanics.test_setting_changes_engine_compatibility
E
E This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch/cuda/__init__.py:579: RuntimeError
|
|
Run Script:
pytorch/tensorrt/tests/py/dynamo/tests/py/dynamo/lowering/test_constant_fold_exclusions.py#L76
TestConstantFoldExclusionMechanics.test_registered_rule_preserves_an_arbitrary_node
self = <dynamo.lowering.test_constant_fold_exclusions.TestConstantFoldExclusionMechanics testMethod=test_registered_rule_preserves_an_arbitrary_node>
def test_registered_rule_preserves_an_arbitrary_node(self):
gm, custom_node = self._custom_graph()
mark_constant_fold_exclusions(gm)
self.assertEqual(
custom_node.meta[CONSTANT_FOLD_EXCLUSION_META_KEY], {self.rule_id}
)
> constant_fold(gm, CompilationSettings())
lowering/test_constant_fold_exclusions.py:83:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
<string>:14: in __init__
???
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch_tensorrt/dynamo/_defaults.py:111: in default_device
return Device(gpu_id=torch.cuda.current_device())
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch/cuda/__init__.py:1267: in current_device
_lazy_init()
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
def _lazy_init():
global _initialized, _queued_calls
if is_initialized() or hasattr(_tls, "is_initializing"):
return
with _initialization_lock:
# We be double-checked locking, boys! This is OK because
# the above test was GIL protected anyway. The inner test
# is for when a thread blocked on some other thread which was
# doing the initialization; when they get the lock, they will
# find there is nothing left to do.
if is_initialized():
return
# It is important to prevent other threads from entering _lazy_init
# immediately, while we are still guaranteed to have the GIL, because some
# of the C calls we make below will release the GIL
if _is_in_bad_fork():
> raise RuntimeError(
"Cannot re-initialize CUDA in forked subprocess. To use CUDA with "
"multiprocessing, you must use the 'spawn' start method"
)
E RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method
E
E To execute this test, run the following from the base repo dir:
E python test_constant_fold_exclusions.py TestConstantFoldExclusionMechanics.test_registered_rule_preserves_an_arbitrary_node
E
E This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch/cuda/__init__.py:579: RuntimeError
|
|
Run Script:
pytorch/tensorrt/tests/py/dynamo/tests/py/dynamo/lowering/test_constant_fold_exclusions.py#L58
TestConstantFoldExclusionMechanics.test_old_serialized_setting_defaults_to_no_disabled_rules
self = <dynamo.lowering.test_constant_fold_exclusions.TestConstantFoldExclusionMechanics testMethod=test_old_serialized_setting_defaults_to_no_disabled_rules>
def test_old_serialized_setting_defaults_to_no_disabled_rules(self):
> state = CompilationSettings().__dict__.copy()
lowering/test_constant_fold_exclusions.py:59:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
<string>:14: in __init__
???
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch_tensorrt/dynamo/_defaults.py:111: in default_device
return Device(gpu_id=torch.cuda.current_device())
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch/cuda/__init__.py:1267: in current_device
_lazy_init()
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
def _lazy_init():
global _initialized, _queued_calls
if is_initialized() or hasattr(_tls, "is_initializing"):
return
with _initialization_lock:
# We be double-checked locking, boys! This is OK because
# the above test was GIL protected anyway. The inner test
# is for when a thread blocked on some other thread which was
# doing the initialization; when they get the lock, they will
# find there is nothing left to do.
if is_initialized():
return
# It is important to prevent other threads from entering _lazy_init
# immediately, while we are still guaranteed to have the GIL, because some
# of the C calls we make below will release the GIL
if _is_in_bad_fork():
> raise RuntimeError(
"Cannot re-initialize CUDA in forked subprocess. To use CUDA with "
"multiprocessing, you must use the 'spawn' start method"
)
E RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method
E
E To execute this test, run the following from the base repo dir:
E python test_constant_fold_exclusions.py TestConstantFoldExclusionMechanics.test_old_serialized_setting_defaults_to_no_disabled_rules
E
E This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch/cuda/__init__.py:579: RuntimeError
|
|
Run Script:
pytorch/tensorrt/tests/py/dynamo/tests/py/dynamo/lowering/test_constant_fold_exclusions.py#L47
TestConstantFoldExclusionMechanics.test_disabled_rules_default_to_empty
self = <dynamo.lowering.test_constant_fold_exclusions.TestConstantFoldExclusionMechanics testMethod=test_disabled_rules_default_to_empty>
def test_disabled_rules_default_to_empty(self):
> self.assertEqual(CompilationSettings().disabled_constant_fold_exclusions, set())
lowering/test_constant_fold_exclusions.py:48:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
<string>:14: in __init__
???
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch_tensorrt/dynamo/_defaults.py:111: in default_device
return Device(gpu_id=torch.cuda.current_device())
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch/cuda/__init__.py:1267: in current_device
_lazy_init()
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
def _lazy_init():
global _initialized, _queued_calls
if is_initialized() or hasattr(_tls, "is_initializing"):
return
with _initialization_lock:
# We be double-checked locking, boys! This is OK because
# the above test was GIL protected anyway. The inner test
# is for when a thread blocked on some other thread which was
# doing the initialization; when they get the lock, they will
# find there is nothing left to do.
if is_initialized():
return
# It is important to prevent other threads from entering _lazy_init
# immediately, while we are still guaranteed to have the GIL, because some
# of the C calls we make below will release the GIL
if _is_in_bad_fork():
> raise RuntimeError(
"Cannot re-initialize CUDA in forked subprocess. To use CUDA with "
"multiprocessing, you must use the 'spawn' start method"
)
E RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method
E
E To execute this test, run the following from the base repo dir:
E python test_constant_fold_exclusions.py TestConstantFoldExclusionMechanics.test_disabled_rules_default_to_empty
E
E This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch/cuda/__init__.py:579: RuntimeError
|
|
Run Script:
pytorch/tensorrt/tests/py/dynamo/tests/py/dynamo/lowering/test_constant_fold_exclusions.py#L86
TestConstantFoldExclusionMechanics.test_disabled_rule_revokes_an_existing_mark
self = <dynamo.lowering.test_constant_fold_exclusions.TestConstantFoldExclusionMechanics testMethod=test_disabled_rule_revokes_an_existing_mark>
def test_disabled_rule_revokes_an_existing_mark(self):
gm, custom_node = self._custom_graph()
custom_node.meta[CONSTANT_FOLD_EXCLUSION_META_KEY] = {self.rule_id}
mark_constant_fold_exclusions(
gm,
> CompilationSettings(disabled_constant_fold_exclusions={self.rule_id}),
)
lowering/test_constant_fold_exclusions.py:92:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
<string>:14: in __init__
???
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch_tensorrt/dynamo/_defaults.py:111: in default_device
return Device(gpu_id=torch.cuda.current_device())
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch/cuda/__init__.py:1267: in current_device
_lazy_init()
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
def _lazy_init():
global _initialized, _queued_calls
if is_initialized() or hasattr(_tls, "is_initializing"):
return
with _initialization_lock:
# We be double-checked locking, boys! This is OK because
# the above test was GIL protected anyway. The inner test
# is for when a thread blocked on some other thread which was
# doing the initialization; when they get the lock, they will
# find there is nothing left to do.
if is_initialized():
return
# It is important to prevent other threads from entering _lazy_init
# immediately, while we are still guaranteed to have the GIL, because some
# of the C calls we make below will release the GIL
if _is_in_bad_fork():
> raise RuntimeError(
"Cannot re-initialize CUDA in forked subprocess. To use CUDA with "
"multiprocessing, you must use the 'spawn' start method"
)
E RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method
E
E To execute this test, run the following from the base repo dir:
E python test_constant_fold_exclusions.py TestConstantFoldExclusionMechanics.test_disabled_rule_revokes_an_existing_mark
E
E This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0
/opt/python/cp310-cp310/lib/python3.10/site-packages/torch/cuda/__init__.py:579: RuntimeError
|
|
Run Script:
pytorch/tensorrt/tests/py/dynamo/lowering/test_aten_lowering_passes.py#L218
TestComplexSubgraph.test_complex_subgraph
AssertionError: Scalars are not equal!
Expected 0 but got 2.
Absolute difference: 2
Relative difference: inf
The following unexpected ops were encountered: {<OpOverload(op='aten.view_as_complex', overload='default')>, <OpOverload(op='aten.view_as_real', overload='default')>}
To execute this test, run the following from the base repo dir:
python test_aten_lowering_passes.py TestComplexSubgraph.test_complex_subgraph
This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0
|
|
Complete job
Node.js 20 is deprecated. The following actions target Node.js 20 but are being forced to run on Node.js 24: ./test-infra/.github/actions/setup-ssh, actions/checkout@v4, conda-incubator/setup-miniconda@v3.1.1, pmeier/pytest-results-action@v0.5.0. For more information see: https://github.blog/changelog/2025-09-19-deprecation-of-node-20-on-github-actions-runners/
|
|
Run Script:
pytorch/tensorrt/tests/py/dynamo/<string>#L6
RuntimeSettings is only honored on TRT-RTX builds; constructing it on regular TensorRT has no effect.
|
|
Run Script:
pytorch/tensorrt/tests/py/dynamo/<string>#L6
RuntimeSettings is only honored on TRT-RTX builds; constructing it on regular TensorRT has no effect.
|
|
Run Script:
pytorch/tensorrt/tests/py/dynamo/<string>#L6
RuntimeSettings is only honored on TRT-RTX builds; constructing it on regular TensorRT has no effect.
|
|
Run Script:
pytorch/tensorrt/tests/py/dynamo/<string>#L6
RuntimeSettings is only honored on TRT-RTX builds; constructing it on regular TensorRT has no effect.
|
|
Run Script:
pytorch/tensorrt/tests/py/dynamo/<string>#L6
RuntimeSettings is only honored on TRT-RTX builds; constructing it on regular TensorRT has no effect.
|
|
Run Script:
../../../../../../../../opt/python/cp310-cp310/lib/python3.10/site-packages/torch/jit/_script.py#L1491
`torch.jit.script` is deprecated. Please switch to `torch.compile` or `torch.export`.
|
|
Run Script:
../../../../../../../../opt/python/cp310-cp310/lib/python3.10/site-packages/torch/jit/_script.py#L1491
`torch.jit.script` is deprecated. Please switch to `torch.compile` or `torch.export`.
|
|
Run Script:
../../../../../../../../opt/python/cp310-cp310/lib/python3.10/site-packages/torch/jit/_script.py#L1491
`torch.jit.script` is deprecated. Please switch to `torch.compile` or `torch.export`.
|
|
Run Script:
../../../../../../../../opt/python/cp310-cp310/lib/python3.10/site-packages/torch/jit/_script.py#L1491
`torch.jit.script` is deprecated. Please switch to `torch.compile` or `torch.export`.
|
|
Run Script:
../../../../../../../../opt/python/cp310-cp310/lib/python3.10/site-packages/torch/jit/_script.py#L1491
`torch.jit.script` is deprecated. Please switch to `torch.compile` or `torch.export`.
|
|
Run ./test-infra/.github/actions/setup-binary-builds
WARNING conda.conda_pypi.main:notify_externally_managed_future(156):
Did you know? You can install many PyPI packages with conda
using the conda-pypi beta. Get started:
https://docs.conda.io/projects/conda/en/stable/new-features.html
|
|
Run ./test-infra/.github/actions/setup-binary-builds
WARNING conda.cli.condarc:set_key(484): Key auto_activate_base is an alias of auto_activate; setting value with latter
|
|
Run ./test-infra/.github/actions/setup-binary-builds
The 'defaults' channel might have been added implicitly. If this is intentional, add 'defaults' to the 'channels' list. Otherwise, consider setting 'conda-remove-defaults' to 'true'.
|
background
wait
wait-all
cancel
parallel
Loading