All pull requests

Pull requests

• Fix flakiness in the ReadRemoteDuplicateFails test.

  #1028·copybara-service[bot] opened Sep 23, 2026·

  ·
  1

• Remove worker_rpc_client parameter and encapsulate WorkerRpcClient in JobEntity

  #1027·copybara-service[bot] opened Sep 23, 2026·

  ·

• Park pulls awaiting registration instead of holding a server thread

  #1026·jndevanshu opened Sep 23, 2026·

  ·

• Add GrpcTransportAdapter with PushBlocks support

  #1025·copybara-service[bot] opened Sep 23, 2026·

  ·

• Optimize tree broadcast hop pre-computation, layer-scoped D2H/H2D, and relay coalescing

  #1024·copybara-service[bot] opened Sep 23, 2026·

  ·

• Bound the consumer pull handshake and stop it holding a pool worker

  pull ready
  #1023·jndevanshu opened Sep 23, 2026·

  ·
  1

  

• Migrate reshard control plane to ControlPipe and remove framed_rpc

  #1017·copybara-service[bot] opened Sep 23, 2026·

  ·

• [tpu-raiden] Implement eviction callback notification upon host LRU cache eviction.

  #1009·copybara-service[bot] opened Sep 22, 2026·

  ·
  1

• Settle zero-block/zero-layer receives and ready sessions in WaitForPendingWork

  #995·copybara-service[bot] opened Sep 21, 2026·

  ·

• [Raiden/FaultTolerance] Support chunk header version compatibility validation and version-dependent parsing

  #993·copybara-service[bot] opened Sep 21, 2026·

  ·
  1

• Add peregrine TransportMetrics and TransportAdapter::GetTransportMetrics().

  #960·copybara-service[bot] opened Sep 18, 2026·

  ·
  1

• [gmon] Add Monarch retention rules for P2P transfer latency and PCIe byte metrics.

  #959·copybara-service[bot] opened Sep 18, 2026·

  ·

• Declare and instrument PCIe H2D and D2H byte counter and per-device label telemetry metrics:

  #950·copybara-service[bot] opened Sep 18, 2026·

  ·

• Add TdsKVBackend storage driver and wire TDS backend into KVCacheManagerBase and KVCacheStore.

  #948·copybara-service[bot] opened Sep 18, 2026·

  ·
  1

• Internal build configuration cleanup for torch_tpu src/ layout

  #938·copybara-service[bot] opened Sep 17, 2026·

  ·

• Add PyTorch support to TPU Raiden H2D/D2H presubmit performance gate for GitHub Actions.

  #925·copybara-service[bot] opened Sep 14, 2026·

  ·
  1

• Add torch_d2d_read_benchmark_runner.py under examples/microbenchmarks/ providing PyTorch parity with jax_d2d_read_benchmark_runner.py:

  #924·copybara-service[bot] opened Sep 14, 2026·

  ·
  1

• ci: build the release matrix from the frameworks input

  #917·yinlin09 opened Sep 13, 2026·

  ·

• OptionA experimenting

  #898·wenjung2007 opened Sep 10, 2026·

  ·
  2

• Implement exponential backoff circuit breaker for failing producer peers in tpu_sync.

  #887·copybara-service[bot] opened Sep 8, 2026·

  ·
  1

• Chunk vectored socket I/O at IOV_MAX instead of rejecting large iovec spans

  #885·majunze2001 opened Sep 8, 2026·

  ·
  1

• Implement gRPC-based transfer control plane for KVCacheManagerWithTransfer.

  #881·copybara-service[bot] opened Sep 8, 2026·

  ·

• Add torch_dma_kv_cache_benchmark.py under examples/microbenchmarks/ providing PyTorch parity with jax_dma_kv_cache_benchmark.py:

  #871·copybara-service[bot] opened Sep 6, 2026·

  ·
  1

• Instrument weight synchronization components with coarse-grained RAIDEN_TRACE annotations.

  #860·copybara-service[bot] opened Sep 5, 2026·

  ·

• Forward skip_tiling across the controller boundary

  pull ready
  #857·lokic233 opened Sep 4, 2026·

  ·
  1