Skip to content
Navigation Menu
Sign in
Products
Platform
GitHub
Repository hosting, code review, and collaboration
Copilot
AI coding agent for tasks, projects, and automated work
Build & Ship
Actions
CI/CD and automation
Code Quality
Quality analysis, fixes, and enforcement
Codespaces
Cloud development environments
Secure
Code Security
Vulnerability detection and remediation
Secret Protection
Secret scanning and leak prevention
All Features
Developers
Learn
Documentation
Training
Certifications
Build
Build with Copilot
Customize Copilot
Integrations
Explore
Trending repos
GitHub Sponsors
Developer articles
Changelog
Changelog
Your source for the latest features, improvements, and fixes across GitHub products, all in one place.
Businesses
By Stage
Enterprises
Team
Startups
Nonprofits
By Industry
Healthcare
Financial services
Manufacturing
Government
See all industries
Evaluate GitHub
Why GitHub
Customer Stories
Business insights
Trust Center
Add-ons
GitHub Advanced Security
Enterprise-grade code security
Copilot for business
Enterprise-grade AI features
Premium Support
Enterprise-grade 24/7 support
Resources
Learn & Connect
Documentation
Resource center
Blog
Events & webinars
Community Forum
Customer Stories
Integrations
Support
Customer support
Training
GitHub Status
Company
About
Partners
Changelog
GitHub Universe
GitHub Universe
The event for developers, by developers — talks, workshops, and what’s next for GitHub.
Pricing
Search
/
Sign in
Sign up
Appearance settings
You signed in with another tab or window.
Reload
to refresh your session.
You signed out in another tab or window.
Reload
to refresh your session.
You switched accounts on another tab or window.
Reload
to refresh your session.
Dismiss alert
{{ message }}
openxla
tokamax
Repository navigation
Code
Issues
22
(22)
Pull requests
145
(145)
Actions
Projects
Security and quality
Insights
More
items
All pull requests
Labels
9
(9)
Milestones
0
(0)
New pull request
Search pull requests
is
:
pr
state
:
open
is:pr state:open
Clear filter
Search
Pull requests
Open
145
(145)
Closed
1,895
(1,895)
Author
Label
Projects
Milestones
Reviews
Assignee
Sort by
Newest
descending
Sort…
Sort by Newest descending
More items
Comfortable display density
Compact display density
Deduplicate explicit-VJP forward tests and fast-fail unsupported
#2069
·
copybara-service[bot]
opened
Nov 19, 2026
Bot
·
·
1
Blackwell (
SM100
) FP8xINT4
ragged_dot
heuristics previously capped
block_k
at 256, disabled 2-CTA
tcgen05
execution, and rescaled TMEM accumulators on every K-tile iteration. That schedule underutilized 2-CTA tensor cores and spent warpgroup cycles on redundant per-tile scale loads and accumulator rescales during decode.
#2068
·
copybara-service[bot]
opened
Nov 19, 2026
Bot
·
·
Copybara import of the project:
#2067
·
copybara-service[bot]
opened
Nov 19, 2026
Bot
·
·
Add an "Adding an Op" guide to the docs.
#2066
·
copybara-service[bot]
opened
Nov 19, 2026
Bot
·
·
Add experimental
fused_qkv_proj
Mosaic TPU kernels and
batched_rpa
pre-packed I/O options.
#2064
·
copybara-service[bot]
opened
Nov 18, 2026
Bot
·
·
1
routing MoE: Optimize group size calculation by using binary search
#2060
·
denis-mil
opened
Nov 18, 2026
·
·
Accept abstract arrays and meshes in routing MoE's bind.
#2059
·
copybara-service[bot]
opened
Nov 18, 2026
Bot
·
·
Internal
#2055
·
copybara-service[bot]
opened
Nov 18, 2026
Bot
·
·
Skip rhs-scale sub-blocking in gmm_v2 when one scale covers the K tile.
#2052
·
copybara-service[bot]
opened
Nov 18, 2026
Bot
·
·
1
Sum the group sizes before group_offset in a separate loop in gmm_v2.
#2045
·
copybara-service[bot]
opened
Nov 17, 2026
Bot
·
·
Benchmark weight-only fp8 GMM v2 on TPUs without fp8 matmuls.
#2036
·
copybara-service[bot]
opened
Nov 17, 2026
Bot
·
·
Overlap routing_moe scatter DMAs with the matmul.
#2035
·
copybara-service[bot]
opened
Nov 17, 2026
Bot
·
·
[Tokamax] Splash attention: optional forward row and sub-tile skip on aligned sliding-window blocks.
#2031
·
copybara-service[bot]
opened
Nov 17, 2026
Bot
·
·
[Tokamax] Splash attention: optional triangle-band-only masking in the dkv kernel on aligned sliding windows.
#2030
·
copybara-service[bot]
opened
Nov 17, 2026
Bot
·
·
[Tokamax] Splash attention: allocate the dkv kernel's dk/dv carriers batched under vmap, with custom_vmap.
#2029
·
copybara-service[bot]
opened
Nov 17, 2026
Bot
·
·
[Tokamax] Splash attention: optional per-q-head VMEM slots for GQA dq/dk/dv in the dkv kernel.
#2028
·
copybara-service[bot]
opened
Nov 17, 2026
Bot
·
·
[Tokamax] Splash attention: keep the static mask schedule with segment ids on near-diagonal masks.
#2027
·
copybara-service[bot]
opened
Nov 17, 2026
Bot
·
·
[Tokamax] Splash attention: trim segment-dead kv bands from full dkv tiles behind a flag.
#2026
·
copybara-service[bot]
opened
Nov 17, 2026
Bot
·
·
[Tokamax] Splash attention: keep each q head's dq resident in VMEM on non-contiguous dq_reduction_steps=1 grids.
#2024
·
copybara-service[bot]
opened
Nov 17, 2026
Bot
·
·
Walk SparseCore top-k slices in whole 8-chunk blocks.
#2023
·
copybara-service[bot]
opened
Nov 17, 2026
Bot
·
·
Canonicalize sm_scale in the batched_rpa autotuning cache key.
#2020
·
copybara-service[bot]
opened
Nov 16, 2026
Bot
·
·
Chunk
array_diff_summary
to avoid host RAM OOMs on large arrays.
#2013
·
copybara-service[bot]
opened
Nov 16, 2026
Bot
·
·
KDA: add opt-in compact all-gather context parallel training
#2009
·
fdz-1999
opened
Nov 15, 2026
Contributor
·
·
Recognize VR200 device names in gpu_utils_test.
#2008
·
copybara-service[bot]
opened
Nov 15, 2026
Bot
·
·
1
Gate XProf re-runs and right-size shards in routing_moe tests.
#2007
·
copybara-service[bot]
opened
Nov 15, 2026
Bot
·
·
1
Previous
1
2
3
4
5
6
Next
You can’t perform that action at this time.