v2.15.0

Issues

• Implemented executorch's multi-optimization profile

  cla signedcomponent: api [C++]component: api [Python]component: corecomponent: dynamocomponent: runtimecomponent: testsdocumentation
  Unread
  pytorch/TensorRT#4441·cehongwang opened Jul 28, 2026 · Updated Sep 29, 2026
  39

• chore: add moe tests

  cla signedcomponent: api [Python]component: conversioncomponent: corecomponent: dynamocomponent: testsForce All Tests[L0+L1+L2]
  Unread
  pytorch/TensorRT#4262·zewenli98 opened May 13, 2026 · Updated Aug 13, 2026
  4

  

• feat: support Global Performance Tuning

  cla signedcomponent: api [Python]component: build systemcomponent: conversioncomponent: corecomponent: dynamocomponent: testsdocumentation
  Unread
  pytorch/TensorRT#4476·zewenli98 opened Aug 11, 2026 · Updated Oct 1, 2026
  7

  

• Fix mutable state leakage from folded constructors

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: loweringcomponent: tests
  Unread
  pytorch/TensorRT#4501·micwill755 opened Aug 17, 2026 · Updated Sep 29, 2026
  8

• fix(dynamo): preserve int64 gather and scatter indices

  ci: fullcla signedcomponent: api [Python]component: conversioncomponent: converterscomponent: corecomponent: dynamocomponent: tests
  Unread
  pytorch/TensorRT#4023·apbose opened Jan 21, 2026 · Updated Oct 1, 2026
  2

• feat: add cutile_op for AOT QDP plugins from cuda tile kernels

  cla signedcomponent: api [Python]component: build systemcomponent: conversioncomponent: corecomponent: dynamocomponent: testsdocumentation
  Unread
  pytorch/TensorRT#4463·bowang007 opened Aug 6, 2026 · Updated Sep 10, 2026

• Cache post-lowering ATen graphs to skip decomp on warm compiles.

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: tests
  Unread
  pytorch/TensorRT#4540·micwill755 opened Aug 20, 2026 · Updated Sep 30, 2026·4/11
  3

• associative_scan lowering

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: loweringcomponent: tests
  Unread
  pytorch/TensorRT#4518·jloftin-nv opened Aug 18, 2026 · Updated Sep 28, 2026
  4

• MLA attention fallback

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: loweringcomponent: tests
  Unread
  pytorch/TensorRT#4517·jloftin-nv opened Aug 18, 2026 · Updated Aug 20, 2026

• MoE histc / _grouped_mm

  cla signedcomponent: api [Python]component: conversioncomponent: converterscomponent: corecomponent: dynamocomponent: tests
  Unread
  pytorch/TensorRT#4514·jloftin-nv opened Aug 18, 2026 · Updated Aug 24, 2026
  2

• Fix strided slice_scatter KV cache updates

  cla signedcomponent: api [Python]component: conversioncomponent: converterscomponent: corecomponent: dynamocomponent: loweringcomponent: tests
  Unread
  pytorch/TensorRT#4550·apbose opened Aug 22, 2026 · Updated Oct 1, 2026

• complex _reshape_copy / _to_copy

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: loweringcomponent: tests
  Unread
  pytorch/TensorRT#4516·jloftin-nv opened Aug 18, 2026 · Updated Aug 22, 2026
  3

• Fix CPU-only import

  cla signedcomponent: api [Python]component: corecomponent: dynamo
  Unread
  pytorch/TensorRT#4504·jloftin-nv opened Aug 18, 2026 · Updated Aug 22, 2026
  2

• Fix noncontiguous AOT inputs

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: tests
  Unread
  pytorch/TensorRT#4509·jloftin-nv opened Aug 18, 2026 · Updated Aug 22, 2026
  2

• Fix cumsum dtype

  cla signedcomponent: api [Python]component: conversioncomponent: converterscomponent: corecomponent: dynamocomponent: tests
  Unread
  pytorch/TensorRT#4511·jloftin-nv opened Aug 18, 2026 · Updated Aug 24, 2026
  4

• feat: add triton_op for AOT QDP plugins from Triton kernels

  cla signedcomponent: api [Python]component: conversioncomponent: corecomponent: dynamocomponent: testsdocumentation
  Unread
  pytorch/TensorRT#4435·bowang007 opened Jul 27, 2026 · Updated Sep 4, 2026
  1

• fix: order the placeholders first before computing serde range constraints

  ci: fullcla signedcomponent: api [Python]component: corecomponent: dynamocomponent: tests
  Unread
  pytorch/TensorRT#4648·tp5uiuc opened Aug 29, 2026 · Updated Sep 29, 2026
  1

  

• fix: keep PyTorch-executed weights on the device when the mutable module offloads

  ci: fullcla signedcomponent: api [Python]component: corecomponent: dynamocomponent: runtimecomponent: tests
  Unread
  pytorch/TensorRT#4647·tp5uiuc opened Aug 29, 2026 · Updated Sep 29, 2026
  2

  

• feat: apply_runtime_settings — permanent runtime config for AOT-loaded engines

  backend: TensorRT-RTXci: fullcla signedcomponent: api [Python]component: corecomponent: dynamocomponent: runtimecomponent: testsdocumentation
  Unread
  pytorch/TensorRT#4570·tp5uiuc opened Aug 24, 2026 · Updated Sep 29, 2026
  

• Apply Torch-TRT decomps during dynamo.trace so compile skips a second AOT retrace.

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: loweringcomponent: runtimecomponent: tests
  Unread
  pytorch/TensorRT#4596·micwill755 opened Aug 26, 2026 · Updated Sep 30, 2026·5/11
  6

• Skip installing large aliased constant-folded weights

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: lowering
  Unread
  pytorch/TensorRT#4531·micwill755 opened Aug 19, 2026 · Updated Sep 30, 2026·2/11
  8

• Batch cheap FX cleanups and defer post_lowering recompile

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: loweringcomponent: tests
  Unread
  pytorch/TensorRT#4532·micwill755 opened Aug 19, 2026 · Updated Sep 30, 2026·3/11
  5

• Skip executing large aliased constant folds that 4531 will not install.

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: loweringcomponent: tests
  Unread
  pytorch/TensorRT#4597·micwill755 opened Aug 26, 2026 · Updated Sep 30, 2026·6/11
  1

• Defer FX graph stringification until DEBUG logging is enabled.

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: loweringcomponent: testscomponent: torch_compile
  Unread
  pytorch/TensorRT#4598·micwill755 opened Aug 26, 2026 · Updated Sep 30, 2026·7/11
  1

• Skip rebuilding FX forward after post-lowering and before adjacency split.

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: loweringcomponent: tests
  Unread
  pytorch/TensorRT#4599·micwill755 opened Aug 26, 2026 · Updated Sep 30, 2026·8/11
  1