[pull] master from tensorflow:master - #8789

Merged

pull[bot] merged 20 commits into

Cache-Cloud:masterfrom

tensorflow:master

Aug 29, 2026

Commits on Aug 28, 2026

Add copyright to XLA's triton patch

rickeylev
authored and
tensorflower-gardener
committed
11fcbbf
[XLA:GPU] Stop lowering scans to the CUB device-scan custom call.

chsigg
authored and
tensorflower-gardener
committed
6d3ce42
Create a heuristic as to whether we should fuse a transpose user into a gemm fusion.

vwbaker
authored and
tensorflower-gardener
committed
314ba25
Revert: [XLA:GPU] Use symmetric memory and peer address lookup for collective kernel thunks.

olegshyshkov
authored and
tensorflower-gardener
committed
c830a86
Rollback due to breakage

junwhanahn
authored and
tensorflower-gardener
committed
8339bc5
Support grouped transposed convolutions in LiteRT-Torch and StableHLO legalization.
tensorflower-gardener
committed
b54a280
[IFRT IR] Do not copy non-donatable inputs on pinned host memory of CPU devices

ICGog
authored and
tensorflower-gardener
committed
b6a3f01
Fix GPU log-determinant calculation for singular matrices

dmiltr3
authored and
tensorflower-gardener
committed
99e23d2
Fix TFLite FlatBuffer SignatureDef indexing and low-bit constant export

majiddadashi
authored and
tensorflower-gardener
committed
a65f6ca
Fix GPU illegal memory access in SparseSegmentSumGrad with empty grad tensor.

dmiltr3
authored and
tensorflower-gardener
committed
d778bbc
move host buffer allocation to HostExecuteStartThunk::Initialize

ermilovmaxim
authored and
tensorflower-gardener
committed
36be3dc
Refactor HLO isolation fusion debugger to use in-memory `expected_literals`
tensorflower-gardener
committed
f89922d
Add 64-bit log metadata to DebugAttributes in hlo.proto
tensorflower-gardener
committed
0adab9a
Handle negative inter_op and intra_op parallelism threads during session and server initialization.

dmiltr3
authored and
tensorflower-gardener
committed
1d31888
Rollback of PR #126109

snnn
authored and
tensorflower-gardener
committed
3c74654
Reduce stack size on c-api DeserializeAndLoad.

pschuh
authored and
tensorflower-gardener
committed
3901de7
Add peak_allocated_bytes stats for pjrt allocator.

bhavya01
authored and
tensorflower-gardener
committed
3a1dd4e
Prepend the standard OpenXLA Apache 2.0 license header to generated

rickeylev
authored and
tensorflower-gardener
committed
229887b
Convert V3 named shardings to V2 in PJRT executable.

Kingston Mandisodza
authored and
tensorflower-gardener
committed
52a89cb

Commits on Aug 29, 2026

Bound integer range in dataflow based input-gen to prevent Inf/NaN.

bhatuzdaname
authored and
tensorflower-gardener
committed
6ca9d84

Navigation Menu

[pull] master from tensorflow:master - #8789

[pull] master from tensorflow:master#8789
pull[bot] merged 20 commits into
Cache-Cloud:masterfrom
tensorflow:master

Commits on Aug 28, 2026

Add copyright to XLA's triton patch

[XLA:GPU] Stop lowering scans to the CUB device-scan custom call.

Create a heuristic as to whether we should fuse a transpose user into a gemm fusion.

Revert: [XLA:GPU] Use symmetric memory and peer address lookup for collective kernel thunks.

Rollback due to breakage

Support grouped transposed convolutions in LiteRT-Torch and StableHLO legalization.

[IFRT IR] Do not copy non-donatable inputs on pinned host memory of CPU devices

Fix GPU log-determinant calculation for singular matrices

Fix TFLite FlatBuffer SignatureDef indexing and low-bit constant export

Fix GPU illegal memory access in SparseSegmentSumGrad with empty grad tensor.

move host buffer allocation to HostExecuteStartThunk::Initialize

Refactor HLO isolation fusion debugger to use in-memory `expected_literals`

Add 64-bit log metadata to DebugAttributes in hlo.proto

Handle negative inter_op and intra_op parallelism threads during session and server initialization.

Rollback of PR #126109

Reduce stack size on c-api DeserializeAndLoad.

Add peak_allocated_bytes stats for pjrt allocator.

Prepend the standard OpenXLA Apache 2.0 license header to generated

Convert V3 named shardings to V2 in PJRT executable.

Commits on Aug 29, 2026

Bound integer range in dataflow based input-gen to prevent Inf/NaN.