-
Notifications
You must be signed in to change notification settings - Fork 147
Pull requests: pytorch/helion
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Fix rope-bwd cudagraph crash and tighten gdn_fwd_h accuracy gate
CLA Signed
This label is managed by the Meta Open Source bot.
#2467
opened May 16, 2026 by
choijon5
Contributor
Loading…
[Pallas] Add test for fused_linear_jsd_fwd autograd path
CLA Signed
This label is managed by the Meta Open Source bot.
[Pallas] remove jax_export_ignore_forward_compatibility setting
CLA Signed
This label is managed by the Meta Open Source bot.
#2452
opened May 15, 2026 by
cota
Collaborator
Loading…
Fused all_gather_fp8_gemm kernel implementation
CLA Signed
This label is managed by the Meta Open Source bot.
#2436
opened May 15, 2026 by
LironKesem
•
Draft
[No Commit] Profiling jagged gdpa on TPU
CLA Signed
This label is managed by the Meta Open Source bot.
#2435
opened May 15, 2026 by
AmesingFlank
Contributor
•
Draft
[Autotuning] Use matmul heuristics as default seeds
CLA Signed
This label is managed by the Meta Open Source bot.
A jagged_gdpa example that works on Pallas TPU
CLA Signed
This label is managed by the Meta Open Source bot.
#2425
opened May 14, 2026 by
AmesingFlank
Contributor
Loading…
{experimental stack not ready for review}[cute] add fp8 dtype mapping and tcgen05 MmaF8F6F4Op capability probe
CLA Signed
This label is managed by the Meta Open Source bot.
[compile] Pin addmm/baddbmm output dtype to self in lowering
CLA Signed
This label is managed by the Meta Open Source bot.
#2420
opened May 13, 2026 by
karthickai
Contributor
•
Draft
[Pallas] jsd if output arity fix
CLA Signed
This label is managed by the Meta Open Source bot.
#2399
opened May 12, 2026 by
thcmbs
Collaborator
Loading…
[Pallas] Slice store values to match clamped Pallas BlockSpec ref shape
CLA Signed
This label is managed by the Meta Open Source bot.
#2398
opened May 12, 2026 by
thcmbs
Collaborator
Loading…
[Pallas] Attention perf: further reduce spillage from pre-loading Q, by loading Q in-loop and not pipelining it
CLA Signed
This label is managed by the Meta Open Source bot.
#2397
opened May 11, 2026 by
AmesingFlank
Contributor
•
Draft
Stop overriding range_num_stages and range_unroll_factor for CUDA IMA workarounds
CLA Signed
This label is managed by the Meta Open Source bot.
[autotune] Add observed heuristic seeds
CLA Signed
This label is managed by the Meta Open Source bot.
[runtime:pallas] migrate to torch_tpu's new Pallas buffer donation API
CLA Signed
This label is managed by the Meta Open Source bot.
#2351
opened May 7, 2026 by
cota
Collaborator
Loading…
Add multi tile loop support to autodiff
CLA Signed
This label is managed by the Meta Open Source bot.
#2338
opened May 7, 2026 by
karthickai
Contributor
•
Draft
[Pallas] Skip factory tensor padding for Pallas backend
CLA Signed
This label is managed by the Meta Open Source bot.
[TPU][Pallas] relax tolerances and fix Pallas autotuning OOM in layer_norm
CLA Signed
This label is managed by the Meta Open Source bot.
#2272
opened May 5, 2026 by
yarongmu-google
Collaborator
•
Draft
A jagged_hstu_attention example that works on Pallas TPU
CLA Signed
This label is managed by the Meta Open Source bot.
#2218
opened May 3, 2026 by
AmesingFlank
Contributor
Loading…
[Pallas] Use LONG_INT_TYPE for jagged offsets in examples and tests
CLA Signed
This label is managed by the Meta Open Source bot.
[Autotuner] Raise default min_improvement_delta to 0.003
CLA Signed
This label is managed by the Meta Open Source bot.
[Pallas] Switch gather to jnp.take_along_axis (for JAX issue filing)
CLA Signed
This label is managed by the Meta Open Source bot.
#2061
opened Apr 20, 2026 by
AmesingFlank
Contributor
•
Draft
Previous Next
ProTip!
Type g i on any issue or pull request to go back to the issue listing page.