-
Notifications
You must be signed in to change notification settings - Fork 32
Pull requests: google/tpu-sync
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Implement exponential backoff circuit breaker for failing producer peers in tpu_sync.
#887
opened Sep 8, 2026 by
copybara-service
Bot
Loading…
Chunk vectored socket I/O at
IOV_MAX instead of rejecting large iovec spans
#885
opened Sep 8, 2026 by
majunze2001
Loading…
Implements raw PJRT device<->host DMA transfers for PyTorch tensors, mirroring
#884
opened Sep 8, 2026 by
copybara-service
Bot
Loading…
Fix ReshardStore to accept the native RaidenId binding.
#883
opened Sep 8, 2026 by
majunze2001
Loading…
Implement gRPC-based transfer control plane for KVCacheManagerWithTransfer.
#881
opened Sep 8, 2026 by
copybara-service
Bot
Loading…
[TPU Raiden] Backend Tier Extension (Part 1/4): Protocols, Interfaces & Abstract Driver Architecture
#880
opened Sep 8, 2026 by
copybara-service
Bot
Loading…
Adopt the torch_tpu tensor_buffer API and drop the nobuilder build exclusions.
#877
opened Sep 7, 2026 by
copybara-service
Bot
Loading…
Integrate PrometheusShmExporter into the RaidenMetricStore facade and add multi-process end-to-end stress tests.
#874
opened Sep 7, 2026 by
copybara-service
Bot
Loading…
Add
torch_dma_kv_cache_benchmark.py under examples/microbenchmarks/ providing PyTorch parity with jax_dma_kv_cache_benchmark.py:
#871
opened Sep 6, 2026 by
copybara-service
Bot
Loading…
Instrument weight synchronization components with coarse-grained
RAIDEN_TRACE annotations.
#860
opened Sep 5, 2026 by
copybara-service
Bot
Loading…
Integrate WeightSynchronizationWorkerService gRPC mode into WeightSynchronizer and RaidenController.
#849
opened Sep 4, 2026 by
copybara-service
Bot
Loading…
Add the LLVM apt source only when the image lacks it
pull ready
#836
opened Sep 1, 2026 by
yinlin09
Collaborator
Loading…
Expand PyTorch KVCacheStore test suite to full parity with JAX:
#830
opened Sep 1, 2026 by
copybara-service
Bot
Loading…
Expand PyTorch WeightSynchronizer test suite to reach parity with JAX:
#828
opened Aug 31, 2026 by
copybara-service
Bot
Loading…
Update TorchTPU KVCacheManager transfer unit test suite and add permutation test cases to reach full parity with JAX:
#824
opened Aug 31, 2026 by
copybara-service
Bot
Loading…
Cut fixed per-request costs out of the reshard control plane
#816
opened Aug 29, 2026 by
copybara-service
Bot
Loading…
Implement ShmCollector to scan, validate, and aggregate telemetry metrics across all active worker memory-mapped files in /dev/shm.
#815
opened Aug 29, 2026 by
copybara-service
Bot
Loading…
Integrate Raiden with Cerebro via UserDefinedOps.
#812
opened Aug 29, 2026 by
copybara-service
Bot
Loading…
Refactor KVCacheManager unit test suite and add transfer test cases:
#810
opened Aug 29, 2026 by
copybara-service
Bot
Loading…
Cut fixed per-request costs out of the reshard control plane
#808
opened Aug 28, 2026 by
yinlin09
Collaborator
Loading…
Previous Next
ProTip!
Adding no:label will show everything without a label.