All pull requests

Pull requests

• Update ExecuTorch pin to 1.6.0.dev20261006

  backend: TensorRTci: nightlycla signedcomponent: api [Python]component: build system
  #4803·pytorchbot opened Oct 7, 2026Collaborator·

  ·

• [DO NOT MERGE] Measure peak container memory on the small A10G pod

  cla signed
  #4802·huydhn opened Oct 7, 2026Contributor·

  ·

• fix constant folding test issues

  ci: fullcla signedcomponent: api [Python]component: corecomponent: dynamocomponent: loweringcomponent: tests
  #4800·lanluo-nvidia opened Oct 7, 2026Collaborator·3/3·7 tasks·

  ·

  · v2.15.0

• cumsum dtype issue fix

  ci: fullcla signedcomponent: tests
  #4799·lanluo-nvidia opened Oct 7, 2026Collaborator·2/3·7 tasks·

  ·

  · v2.15.0

• Grant the Linux wheel builds access to the release runner groups

  ci: fullcla signed
  #4798·huydhn opened Oct 7, 2026Contributor·

  ·

  · v2.15.0
  5

• perf(conversion): hand the built engine's plan to the runtime without copying it

  cla signedcomponent: api [Python]component: conversioncomponent: corecomponent: dynamocomponent: runtimecomponent: testsdocumentation
  #4796·narendasan opened Oct 7, 2026Collaborator·5/5·

  ·

• perf(runtime): don't keep a host copy of the plan once the engine is set up

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: runtimecomponent: tests
  #4795·narendasan opened Oct 7, 2026Collaborator·4/5·

  ·

• perf(runtime): build C++ engines from a view of the plan instead of copies

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: runtimecomponent: tests
  #4794·narendasan opened Oct 7, 2026Collaborator·3/5·

  ·

• perf(dynamo): offload modules to CPU without holding both copies

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: tests
  #4793·narendasan opened Oct 7, 2026Collaborator·2/5·

  ·

• perf(dynamo): return builder heap to the OS after each engine build

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: testsdocumentation
  #4792·narendasan opened Oct 7, 2026Collaborator·1/5·

  ·

• remove python 3.10 support

  ci: fullcla signedcomponent: api [Python]component: build systemcomponent: fxcomponent: testsdocumentationfx
  #4791·lanluo-nvidia opened Oct 7, 2026Collaborator·1/3·7 tasks·

  ·

  · v2.15.0
  4

• upgrade torch_tensorrt to 2.16.0

  cla signedcomponent: api [Python]component: build systemdocumentation
  #4790·lanluo-nvidia opened Oct 7, 2026Collaborator·7 tasks·

  ·

• ix(converters): trunc_div sign overflow gives wrong remainder/fmod for large integers

  cla signedcomponent: api [Python]component: conversioncomponent: converterscomponent: corecomponent: dynamocomponent: tests
  #4786·kodlan opened Oct 6, 2026Contributor·5 of 7 tasks·

  ·
  1

• fix(converters): embedding_bag returns zeros for the last bag with dynamic indices and constant offsets

  cla signedcomponent: api [Python]component: conversioncomponent: converterscomponent: corecomponent: dynamocomponent: tests
  #4785·kodlan opened Oct 6, 2026Contributor·6 of 7 tasks·

  ·
  3

• [MD-TRT] Llama and Qwen export TP examples with KV cache

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: tests
  #4784·apbose opened Oct 6, 2026Collaborator·3 of 4 tasks·

  ·

  · v2.15.0

• fix(partitioning): keep getitem with its multi-output producer when splitting by size

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: tests
  #4783·cehongwang opened Oct 5, 2026Collaborator·5 of 7 tasks·

  ·

• feat(executorch): optional CUDA graph replay in the TensorRT delegate

  ci: fullcla signedcomponent: api [C++]component: api [Python]component: tests
  #4781·shoumikhin opened Oct 5, 2026Contributor·

  ·

• fix(executorch): load TensorRT engines through ExecuTorch's CUDA allocator

  cla signedcomponent: api [C++]component: tests
  #4780·shoumikhin opened Oct 5, 2026Contributor·

  ·

• [MD-TRT] enforce WORLD parent and gate native subgroup tests for now

  cla signedcomponent: api [Python]component: conversioncomponent: converterscomponent: corecomponent: dynamocomponent: runtimecomponent: tests
  #4772·apbose opened Oct 1, 2026Collaborator·

  ·

  · v2.16.0

• fix(dynamo): keep dryrun non-fatal and repair RTX fallback test

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: tests
  #4768·kvnloo opened Oct 1, 2026·

  ·
  2

• feat(runtime)[DRAFT]: add global engine profiling controls

  cla signedcomponent: corecomponent: runtimecomponent: tests
  #4766·apbose opened Oct 1, 2026Collaborator·

  ·

  · v2.16.0

• Add unified ExecuTorch exporter

  cla signedcomponent: api [C++]
  #4763·micwill755 opened Sep 30, 2026Collaborator·13/13·

  ·

• Add PI0.5 action delegate

  cla signedcomponent: api [C++]component: tests
  #4762·micwill755 opened Sep 30, 2026Collaborator·12/13·5 of 7 tasks·

  ·

• Add language decode delegate

  cla signedcomponent: api [C++]component: tests
  #4760·micwill755 opened Sep 30, 2026Collaborator·11/13·5 of 7 tasks·

  ·

• Add language prefill delegate

  cla signedcomponent: api [C++]component: tests
  #4759·micwill755 opened Sep 30, 2026Collaborator·10/13·5 of 7 tasks·

  ·