All pull requests

Pull requests

• docs: CPU compilation and resource partitioning in the resource management guide

  cla signeddocumentation
  #4858·cehongwang opened Oct 13, 2026Collaborator·3 of 7 tasks·

  ·

• fix: check_output_equal only compared the first element of lists/tuples

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: tests
  #4857·Sujendra-23 opened Oct 13, 2026·3 of 7 tasks·

  ·

• docs: remove stale runtime_cache_path compile() setting

  documentation
  #4856·Sujendra-23 opened Oct 13, 2026·3 of 7 tasks·

  ·
  1

• feat(dynamo): resource_partition_engine_cap - a fixed per-engine size for the resource partitioner

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: testsdocumentation
  #4855·cehongwang opened Oct 12, 2026Collaborator·6 of 7 tasks·

  ·

• feat(dynamo): map_weights - compile from memory-mapped checkpoint weights

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: tests
  #4854·cehongwang opened Oct 12, 2026Collaborator·6 of 7 tasks·

  ·

• feat(dynamo): release model weights once their last engine is built

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: tests
  #4853·cehongwang opened Oct 12, 2026Collaborator·6 of 7 tasks·

  ·

• Resolve ExecuTorch without a published native companion

  ci: fullcla signedcomponent: api [Python]component: build systemcomponent: tests
  #4851·lanluo-nvidia opened Oct 12, 2026Collaborator·1/2·

  ·

  · v2.15.0

• fix(converters): true division of integer tensors truncates for int64 and scalar divisors

  cla signedcomponent: api [Python]component: conversioncomponent: converterscomponent: corecomponent: dynamocomponent: tests
  #4850·usmanmateen opened Oct 12, 2026·5 of 7 tasks·

  ·
  1

• feat(executorch): optional .ptd file for TensorRT engines, refuse engines without weights

  cla signedcomponent: api [C++]component: api [Python]component: tests
  #4848·shoumikhin opened Oct 11, 2026Contributor·

  ·

• Upgrade TensorRT from 11.3 to 11.4

  cla signedcomponent: api [Python]component: build systemcomponent: tests
  #4845·lanluo-nvidia opened Oct 10, 2026Collaborator·2/2·

  ·

• docs: add an ExecuTorch deployment guide

  cla signeddocumentation
  #4839·shoumikhin opened Oct 9, 2026Contributor·

  ·
  1

• Add Windows support for ExecuTorch

  ci: fullcla signedcomponent: api [C++]component: api [Python]component: build systemcomponent: tests
  #4835·lanluo-nvidia opened Oct 9, 2026Collaborator·

  ·

  · v2.15.0

• fix(executorch): allow CUDA graph replay on shared activation scratch

  ci: fullcla signedcomponent: api [C++]component: tests
  #4833·shoumikhin opened Oct 9, 2026Contributor·

  ·

• [MD-TRT] Support subgroup parent communicators for native collectives

  cla signedcomponent: api [Python]component: conversioncomponent: converterscomponent: corecomponent: dynamocomponent: runtimecomponent: tests
  #4829·apbose opened Oct 8, 2026Collaborator·2/2·

  ·

• perf(executorch): store each engine once, as named data, when saving a .pte

  cla signedcomponent: api [C++]component: api [Python]component: tests
  #4821·cehongwang opened Oct 8, 2026Collaborator·

  ·

• perf(partitioning): free memory before the resource partitioner reads RSS

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: tests
  #4820·cehongwang opened Oct 8, 2026Collaborator·5 of 7 tasks·

  ·

• fix(converters): let converters declare uint8 support and reject uint8 tensors otherwise

  cla signedcomponent: api [Python]component: conversioncomponent: corecomponent: dynamocomponent: tests
  #4819·narendasan opened Oct 7, 2026Collaborator·

  ·

• fix(executorch): warn when a saved program calls operators the runtime cannot run

  cla signedcomponent: api [Python]component: tests
  #4818·narendasan opened Oct 7, 2026Collaborator·

  ·

• fix(executorch): save programs that keep a view outside the TensorRT engines

  cla signedcomponent: api [Python]component: tests
  #4817·narendasan opened Oct 7, 2026Collaborator·

  ·

• fix(executorch): give the lowered program host copies of its CUDA constants

  cla signedcomponent: api [Python]component: tests
  #4816·narendasan opened Oct 7, 2026Collaborator·

  ·

• fix(dynamo): give the float32 cast that truncate_double inserts a value

  cla signedcomponent: api [Python]component: conversioncomponent: corecomponent: dynamocomponent: tests
  #4815·narendasan opened Oct 7, 2026Collaborator·

  ·

• fix(lowering): normalize a negative index in the select_scatter decomposition

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: loweringcomponent: tests
  #4814·narendasan opened Oct 7, 2026Collaborator·

  ·

• fix(converters): accept constant query, key and value in attention converters

  cla signedcomponent: api [Python]component: conversioncomponent: converterscomponent: corecomponent: dynamocomponent: tests
  #4813·narendasan opened Oct 7, 2026Collaborator·

  ·

• fix(converters): clamp out-of-range static slice bounds instead of wrapping them

  cla signedcomponent: api [Python]component: conversioncomponent: converterscomponent: corecomponent: dynamocomponent: tests
  #4812·narendasan opened Oct 7, 2026Collaborator·

  ·

• fix(dynamo): save compiled modules whose single output is a bare graph node

  cla signedcomponent: api [Python]component: corecomponent: dynamocomponent: tests
  #4811·narendasan opened Oct 7, 2026Collaborator·

  ·