-
Notifications
You must be signed in to change notification settings - Fork 3.2k
Pull requests: jax-ml/jax
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[Mosaic GPU] Disable
mypy
in layout inference files.
#32333
by copybara-service
bot
was merged Oct 3, 2025
Loading…
updated Oct 3, 2025
[Mosaic GPU][NFC] Clean up type checking errors.
#32324
by copybara-service
bot
was merged Oct 3, 2025
Loading…
updated Oct 3, 2025
Remove deprecated non-hermetic CUDA and NCCL repo rules.
#32331
by copybara-service
bot
was merged Oct 3, 2025
Loading…
updated Oct 3, 2025
[Pallas/Mosaic GPU] Extend test coverage for
hopper_mixed_type_matmul_mgpu
.
#32329
by copybara-service
bot
was closed Oct 3, 2025
Loading…
updated Oct 3, 2025
[Mosaic GPU] Signal pass failure in
AssemblyToBinaryPass
if PTX compilation failed.
#32330
by copybara-service
bot
was merged Oct 3, 2025
Loading…
updated Oct 3, 2025
[Pallas][Mosaic GPU] Improve ragged dot kernel for blackwell.
#31708
by copybara-service
bot
was merged Sep 15, 2025
Loading…
updated Oct 3, 2025
[Mosaic TPU] Use integer types for transpose emulation
#32289
by copybara-service
bot
was merged Oct 3, 2025
Loading…
updated Oct 3, 2025
[Mosaic TPU] Support scalar input-output-alias. Relax the condition in
ShapeVerifier::HandleCustomCall
to allow scalar operand live on S(6) while the result has no memory space assigned.
#32116
by copybara-service
bot
was merged Oct 3, 2025
Loading…
updated Oct 3, 2025
Fix out_memory_spaces_flat calculation. It was using in_tree before but should have used out_tree instead.
#32322
by copybara-service
bot
was merged Oct 3, 2025
Loading…
updated Oct 3, 2025
Add experimental higher-order primitive called compute_on2. This should eventually replace the current
compute_on
.
#32319
by copybara-service
bot
was merged Oct 3, 2025
Loading…
updated Oct 3, 2025
Clarify
optimization_barrier
semantics.
#32309
by copybara-service
bot
was merged Oct 2, 2025
Loading…
updated Oct 2, 2025
Remove uses of HloModuleGroup.
#32307
by copybara-service
bot
was closed Oct 2, 2025
Loading…
updated Oct 2, 2025
[Mosaic GPU][NFC] Specify the memory space for two existing tests.
#32250
by copybara-service
bot
was merged Oct 2, 2025
Loading…
updated Oct 2, 2025
[Mosaic GPU] Add SMEM inference rules for
with_transforms
and slice_smem
.
#32249
by copybara-service
bot
was merged Oct 2, 2025
Loading…
updated Oct 2, 2025
Speed up unpacking
.tar.xz
files in the hermetic toolchain by using multithreaded xz
execution.
#32262
by copybara-service
bot
was merged Oct 2, 2025
Loading…
updated Oct 2, 2025
Support collective cancelling in TFRT GPU client.
#30729
by copybara-service
bot
was closed Oct 2, 2025
Loading…
updated Oct 2, 2025
[pmap] Fix up remaining docstring examples and tests.
#32306
by copybara-service
bot
was merged Oct 2, 2025
Loading…
updated Oct 2, 2025
[sparse_bcoo_bcsr] Reduce the number of sampled tests for cpu variants to avoid timeouts.
#32304
by copybara-service
bot
was merged Oct 2, 2025
Loading…
updated Oct 2, 2025
[Pallas/Fuser] Add basic hitype support to fuser.fuse
#32277
by copybara-service
bot
was merged Oct 2, 2025
Loading…
updated Oct 2, 2025
Update test shardings.
#32303
by copybara-service
bot
was merged Oct 2, 2025
Loading…
updated Oct 2, 2025
[Mosaic GPU] Handle SMEM in the
tmem_alloc
layout inference rule.
#32239
by copybara-service
bot
was closed Oct 2, 2025
Loading…
updated Oct 2, 2025
[JAX] Adjusting mesh for allowing 2 cores per chip on V5P
#32298
by copybara-service
bot
was merged Oct 2, 2025
Loading…
updated Oct 2, 2025
Update
rules_ml_toolchain
to newer version with latest updates and SYCL/OneAPI support
#32273
by copybara-service
bot
was merged Oct 2, 2025
Loading…
updated Oct 2, 2025
[Pallas/Mosaic GPU][Docs] Add missing backtick to reference.
#32301
by copybara-service
bot
was merged Oct 2, 2025
Loading…
updated Oct 2, 2025
[pallas:sc]
pltpu.async_copy
can now do dynamically-sized scatters and gathers
#32288
by copybara-service
bot
was merged Oct 2, 2025
Loading…
updated Oct 2, 2025
Previous Next
ProTip!
Find all pull requests that aren't related to any open issues with -linked:issue.