-
Notifications
You must be signed in to change notification settings - Fork 599
Pull requests: AI-Hypercomputer/maxtext
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Add SparseCore collective offloading for MoE with custom VJP for FSDP All-Gather to Reduce-Scatter.
#5125
opened Sep 3, 2026 by
copybara-service
Bot
Loading…
Optimize top-2 expert group score computation in MoE routing.
#5124
opened Sep 3, 2026 by
copybara-service
Bot
Loading…
[megablox] Remove redundant expert-weight transpose copy in unquantized gmm_v2 DLHS
#5123
opened Sep 3, 2026 by
copybara-service
Bot
Loading…
Add flag governing whether to apply Muon to MoE routers.
#5122
opened Sep 3, 2026 by
copybara-service
Bot
Loading…
fix(dependencies): reduce container image vulnerabilities in TPU stable images
#5121
opened Sep 3, 2026 by
khatwanimohit
Collaborator
Loading…
4 tasks done
Support MLPerf training logging compliance for MaxText
gemini-review
#5118
opened Sep 3, 2026 by
zxhe-sean
Collaborator
Loading…
4 tasks done
Implement QK attention head chunking for CSA [Deepseek v4]
#5116
opened Sep 3, 2026 by
octatrifan
Collaborator
•
2/2
•
Draft
4 tasks
Keep qwix out of tpu-inference's fused MoE kernel; pre-quantize its weights
#5114
opened Sep 3, 2026 by
wenxindongwork
Collaborator
•
Draft
3 of 4 tasks
vLLM adapter: index kv_caches through layer_name_to_kvcache_index
#5113
opened Sep 2, 2026 by
wenxindongwork
Collaborator
•
Draft
3 of 4 tasks
feat(rl): plumb GRPO flags to the agentic learner
#5107
opened Sep 2, 2026 by
entrpn
Collaborator
Loading…
4 tasks done
Migrate MaxText training engine checkpointing to Orbax v1 Checkpointer API
#5103
opened Sep 2, 2026 by
copybara-service
Bot
Loading…
Add RL sequence-packing and vLLM block-size knobs
#5100
opened Sep 2, 2026 by
gutianyu-google
•
Draft
4 tasks done
Defer the data-parallel gradient all-reduce, and make ZeRO-1 work, in MaxTextTrainingEngine
#5099
opened Sep 2, 2026 by
NuojCheng
Collaborator
Loading…
4 tasks done
Test Fused GDN Backward Pass Kernel w/ MaxText
#5098
opened Sep 2, 2026 by
Rohan-Bierneni
Collaborator
•
Draft
4 tasks done
Training engine: implement eval_step API
#5097
opened Sep 2, 2026 by
SurbhiJainUSC
Collaborator
Loading…
4 tasks done
Update pyconfig to get config defaults from types.py instead of base.yml
#5096
opened Sep 1, 2026 by
bvandermoon
Collaborator
Loading…
4 tasks done
Implement continuous stream chunking for pre-tokenized C4 MLPerf dataset
#5095
opened Sep 1, 2026 by
snehalv2002
Collaborator
•
Draft
4 tasks
Fixed e2e test for Qwen3.5-35b-a3b
pull ready
#5094
opened Sep 1, 2026 by
AusarYao28
Collaborator
Loading…
4 tasks done
[WIP] [Trellis] Weight Conversion that is compatible with Raiden
#5089
opened Sep 1, 2026 by
YixuanWang-99
Collaborator
Loading…
4 tasks
Make tensorflow optional in requirements
#5086
opened Sep 1, 2026 by
aireenmei
Collaborator
Loading…
4 tasks done
Previous Next
ProTip!
Type g p on any issue or pull request to go back to the pull request listing page.