All pull requests

Pull requests

• Fix: VRAM leak after nvshmemx_buffer_unregister_symmetric when P2P is enabled

  #111·gzshan opened Sep 23, 2026·

  ·

• Initialize proxy global-exit request state

  #110·zianglih opened Sep 21, 2026·

  ·
  1

• init: release the init stream and handle arrays at finalize

  #109·0z5a opened Sep 18, 2026·

  ·
  1

• nvshmem4py: fix the ring expectation in test_put_on_tensor

  #106·0z5a opened Sep 18, 2026·

  ·
  1

• perftest: add an EP-shaped small-message RMA benchmark

  #105·0z5a opened Sep 18, 2026·

  ·
  1

• init: create state->my_stream once

  #104·0z5a opened Sep 18, 2026·

  ·
  1

  2

• nvshmem4py: add bfloat16 to the CuTe device-side dispatch

  #102·zkyue opened Sep 1, 2026Contributor·

  ·
  8

• nvshmem4py: support torch.float8_e4m3fn / float8_e5m2 tensor allocation

  #101·zkyue opened Sep 1, 2026Contributor·

  ·
  1

• libfabric: fix multi-NIC RX imbalance for EFA transport

  #76·crazyguitar opened Mar 25, 2026·

  ·

• fix SIGSEGV when release resources

  #52·alpha-baby opened Jan 5, 2026·

  ·

• Fix CUDA_HOME cmake arg handling in devicelib

  #49·kylosus opened Jan 2, 2026·

  ·

• [Bugfix][feat]: add support for fp16 and bf16 dtype aliases for collective operations.

  #46·foraxe opened Dec 12, 2025·

  ·
  1

• transport/ibrc: Make srq configurable in IBRC

  #41·GuangguanWang opened Dec 1, 2025·

  ·

• transport: Replace std::vector with arrays in proxy put-signal API

  #38·a-szegel opened Nov 26, 2025Contributor·

  ·

• Inherit traffic class from global_tclass for IBGDA traffic

  #25·Alvincyq opened Nov 10, 2025·

  ·

• cmake: Fix conflicting optimization flags in debug builds

  #15·a-szegel opened Oct 11, 2025Contributor·

  ·