1 Release published by 1 person

• v2.6.1 Torch-TensorRT v2.6.1

  published Jun 3, 2025

45 Pull requests merged by 9 people

• disable aoti format in windows

  #3632 merged Jun 28, 2025

• disable test on aarch64 for now

  #3623 merged Jun 27, 2025

• fix ts fe test error.

  #3619 merged Jun 26, 2025

• fix: Fix unbacked sym int not found issue

  #3617 merged Jun 26, 2025

• remove debug flag

  #3618 merged Jun 25, 2025

• a few CI changes

  #3612 merged Jun 25, 2025

• TorchTensorRTModule Serialization Fix

  #3572 merged Jun 24, 2025

• fix cross compilation test bug

  #3609 merged Jun 24, 2025

• py39 does not like | E TypeError: unsupported operand type(s) for |: 'type' and 'EnumMeta'

  #3611 merged Jun 24, 2025

• fix: pre-commit issues

  #3603 merged Jun 19, 2025

• fix: add prefix in hierarchical_partitioner_example

  #3607 merged Jun 18, 2025

• debug flag is deprecated, remove it so that test won't complain

  #3610 merged Jun 18, 2025

• fix enabled_precisions error in test cases

  #3606 merged Jun 18, 2025

• skip flashinfer-python for py3.9 due to upstream error

  #3605 merged Jun 18, 2025

• feat: Saving modules using the AOTI format

  #3567 merged Jun 18, 2025

• fix dynamo converter test case failure

  #3594 merged Jun 18, 2025

• feat: Hierarchical Partitioner to support multi-backends

  #3539 merged Jun 18, 2025

• Import dllist only on linux

  #3592 merged Jun 17, 2025

• test failed but displayed as green

  #3599 merged Jun 17, 2025

• Changed weight map to tensor and fix the refit bug

  #3573 merged Jun 17, 2025

• remove torchvision dependency from build, optional for test

  #3598 merged Jun 17, 2025

• feat: working uv pyproject.toml

  #3597 merged Jun 17, 2025

• Fix: fix aot plugin example docstring issue

  #3595 merged Jun 17, 2025

• Verify C++ tests, fix cuda graphs union issue

  #3589 merged Jun 16, 2025

• fix main test failure bug

  #3590 merged Jun 16, 2025

• FX graph visualization

  #3528 merged Jun 13, 2025

• Added flux demo

  #3418 merged Jun 13, 2025

• fix: Fix a perf regression due to weights being ITensors

  #3568 merged Jun 13, 2025

• Add fp4 support

  #3532 merged Jun 13, 2025

• add --no-deps for tests/py/requirements.txt

  #3569 merged Jun 12, 2025

• fix: Fix constant folding failure due to modelopt

  #3565 merged Jun 11, 2025

• Cross compile guard

  #3486 merged Jun 11, 2025

• Upgrade to TensorRT 10.11

  #3557 merged Jun 8, 2025

• fix int8/fp8 constant folding issue

  #3543 merged Jun 6, 2025

• fix the jetson nightly build check bug

  #3552 merged Jun 6, 2025

• Changed the debug setting

  #3551 merged Jun 6, 2025

• fix: handle device in the same way as dtype in aten.full_like decomposition

  #3538 merged Jun 5, 2025

• Publish jetson wheel to pytorch nightly index

  #3550 merged Jun 4, 2025

• feat: TensorRT AOT Plugin

  #3504 merged Jun 4, 2025

• Add jetson build on CI

  #3524 merged Jun 3, 2025

• cherry pick 3505(windows driver upgrade) to release2.6.1

  #3547 merged Jun 3, 2025

• nvidia-modelopt dependency fix

  #3544 merged Jun 2, 2025

• update version to 2.6.1

  #3545 merged Jun 2, 2025

• fix the build issue for patch2.6.1

  #3542 merged May 30, 2025

• remove breakpoint

  #3540 merged May 30, 2025

10 Pull requests opened by 5 people

• add a quick fail fast feed-back actions before the full tests is triggered

  #3541 opened May 30, 2025

• fix: replace add_identity by add_cast for type cast

  #3563 opened Jun 9, 2025

• adding rotary embedding example, with graph rewrite for complex subgraph

  #3570 opened Jun 13, 2025

• Cpu memory optimization

  #3602 opened Jun 17, 2025

• Remove Bazel files from wheel

  #3615 opened Jun 25, 2025

• add weekly cron job to upgrade uv lock

  #3616 opened Jun 25, 2025

• Refit debug patch

  #3620 opened Jun 25, 2025

• Closed the perf gap of resnet and enabled refit

  #3629 opened Jun 27, 2025

• fix the broken CC0 image link

  #3635 opened Jun 28, 2025

• Temporary fix to workaround the mutable decomposition error.

  #3636 opened Jun 28, 2025

19 Issues closed by 8 people

• 🐛 [Bug] Windows dynamo export serde test failed

  #3630 closed Jun 28, 2025

• 🐛 [Bug] Test failure `tests.py.ts.api.test_classes`

  #3578 closed Jun 28, 2025

• 🐛 [Bug] Reexport test failures

  #3588 closed Jun 26, 2025

• 🐛 [Bug] Using `torch_executed_ops` causes the metadata packing to fail

  #3566 closed Jun 24, 2025

• 🐛 [Bug] tests.py.dynamo.automatic_plugin.test_flashinfer_rmsnorm TypeError: dataclass() got an unexpected keyword argument 'kw_only'

  #3577 closed Jun 19, 2025

• 🐛 [Bug] Refit test failures

  #3587 closed Jun 19, 2025

• 🐛 [Bug] tests/py/dynamo/conversion/test_atan2_aten.py Conversion of function torch._ops.aten.aten::_assert_scalar not currently supported!

  #3581 closed Jun 19, 2025

• 🐛 [Bug] Workflow issue: the tests are failed but showing as green in the workflow UI

  #3593 closed Jun 19, 2025

• 🐛 [Bug] mypy precommit issues

  #3600 closed Jun 19, 2025

• 🐛 [Bug] ModuleNotFoundError: No module named 'dllist'

  #3591 closed Jun 17, 2025

• ❓ [Question] Can I export a serialized engine from Torch-TensorRT targeting TensorRT 10.3.0.26?

  #3571 closed Jun 16, 2025

• 🐛 [Bug] Severe performance regression with 2.7 and 2.8.dev

  #3561 closed Jun 13, 2025

• 🐛 [Bug] AttributeError: '_OpNamespace' 'tensorrt' object has no attribute 'quantize_op'

  #3560 closed Jun 13, 2025

• Constant folding crashes if model opt is not installed

  #3564 closed Jun 11, 2025

• Update the lock file to reflect the torch nightly

  #3553 closed Jun 8, 2025

• 🐛 [Bug] RuntimeError: Unhandled FakeTensor Device Propagation for torchvision.deform_conv2d.default, found two different devices cuda:0, cpu

  #3556 closed Jun 7, 2025

• Please Promote torch_tensorrt 2.6.1 release artifacts

  #3546 closed Jun 3, 2025

• ✨[Feature] Add `pickle_protocol` argument for `torch_tensorrt.save`

  #3294 closed May 30, 2025

• ✨[Feature] Warn users who are building Torch-TensorRT with CUDA 12.6 that they need to use CXX11 ABI

  #3339 closed May 30, 2025

23 Issues opened by 8 people

• 🐛 [Bug] Changing input size would affect the TRT engine size, testing on BERT

  #3634 opened Jun 28, 2025

• 🐛 [Bug] Dynamo core test failed in linux

  #3633 opened Jun 27, 2025

• 🐛 [Bug] Windows dynamo core tests failed

  #3631 opened Jun 27, 2025

• 🐛 [Bug] current BN implementation results in slower performance

  #3628 opened Jun 26, 2025

• 📖 [Story] Performance improvements/gaps in Torch-TensorRT

  #3627 opened Jun 26, 2025

• 🐛 [Bug] Large Accuracy Issue

  #3626 opened Jun 26, 2025

• ✨[Feature] Integrate release pipeline with test pipeline

  #3625 opened Jun 26, 2025

• ✨[Feature] Automate examples verification in the release pipeline

  #3624 opened Jun 26, 2025

• compile the QAT trained model in with TensorRT

  #3622 opened Jun 26, 2025

• error in data types

  #3614 opened Jun 25, 2025

• 🐛 [Bug] Flux perf scripts issue

  #3601 opened Jun 17, 2025

• 🐛 [Bug] torchtrt.dynamo.compile produces nan values

  #3596 opened Jun 16, 2025

• 🐛 [Bug] MutableModule Test Failures

  #3586 opened Jun 14, 2025

• 🐛 [Bug] Weight streaming test failures

  #3585 opened Jun 14, 2025

• 🐛 [Bug] FAILED runtime/test_003_cross_compile_for_windows.py::TestCrossCompileSaveForWindows::test_dynamo_cross_compile_for_windows_multiple_output - Failed: unexpected exception raised: Windows cross compilation feature is not available

  #3584 opened Jun 14, 2025

• 🐛 [Bug] FAILED runtime/test_002_lazy_engine_init.py::TestLazyEngineInit::test_lazy_engine_init_cpp_e2e - AssertionError: False is not true : Resnet18 TRT outputs don't match with the original model. Cosine sim score: nan Threshold: 0.99

  #3583 opened Jun 14, 2025

• 🐛 [Bug] FAILED automatic_plugin/test_automatic_plugin.py::TestAutomaticPlugin::test_mul_plugin_float_0 - AssertionError: Tensor-likes are not close!

  #3582 opened Jun 14, 2025

• 🐛 [Bug] tests.py.dynamo.models.test_dyn_models.test_base_dynamic

  #3580 opened Jun 14, 2025

• 🐛 [Bug] tests.py.dynamo.models.test_dtype_support.TestBF16Support.test_bf16_cpp

  #3579 opened Jun 14, 2025

• 🐛 [Bug] .tests.py.ts.api.test_classes

  #3576 opened Jun 13, 2025

• 🐛 [Bug] Test failure in tests/py/dynamo/conversion/test_abs_aten.py ModuleNotFoundError: No module named 'torch_tensorrt.dynamo.debug'

  #3575 opened Jun 13, 2025

• Is it possible to enable Github Merge Queues for our Repos?

  #3559 opened Jun 6, 2025

• Remove the distributed extra

  #3554 opened Jun 6, 2025

11 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• feat: Refactor LLM model zoo and add KV cache support

  #3527 commented on Jun 27, 2025 • 13 new comments

• TRT-LLM loading mechanism tool

  #3398 commented on Jun 27, 2025 • 5 new comments

• slice scatter support for dynamic cases

  #3513 commented on Jun 13, 2025 • 1 new comment

• AOT plugin: examples with RMSNORM

  #3529 commented on Jun 12, 2025 • 1 new comment

• 🐛 [Bug] Difficulties Quantizing FP16 Models to INT8 Using torch_tensorrt (MLP, CNN, Attention, LSTM, Transformer)

  #3494 commented on May 30, 2025 • 0 new comments

• ❓ [Question] Manually Annotate Quantization Parameters in FX Graph

  #3522 commented on Jun 2, 2025 • 0 new comments

• ✨[Feature] Add FX tests to CI

  #3492 commented on Jun 6, 2025 • 0 new comments

• 🐛 [Bug] AssertionError: end must be an integer

  #3448 commented on Jun 12, 2025 • 0 new comments

• ❓ [Question] Is SAM2 supported when compiling with the Dynamo backend on JetPack 6.1 or 6.2?

  #3478 commented on Jun 28, 2025 • 0 new comments

• Debugging guide for TensorRT

  #3489 commented on Jun 9, 2025 • 0 new comments

• flux fp4 example(WIP)

  #3537 commented on Jun 12, 2025 • 0 new comments