-
Notifications
You must be signed in to change notification settings - Fork 554
Pull requests: AI-Hypercomputer/maxtext
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Separate RL configurations into a dedicated config class
#4431
opened Jul 10, 2026 by
SurbhiJainUSC
Collaborator
•
Draft
4 tasks done
update tpu-inference and vllm versions
#4430
opened Jul 10, 2026 by
A9isha
Collaborator
Loading…
4 tasks done
add XLML tests for DeepSeek V4 Flash
#4426
opened Jul 10, 2026 by
snehalv2002
Collaborator
•
Draft
4 tasks
Enable padded video inputs in the Qwen3* vision encoder
pull ready
#4425
opened Jul 10, 2026 by
hengtaoguo
Collaborator
Loading…
4 tasks done
deprecate version guard in ragged kernels
pull ready
#4423
opened Jul 10, 2026 by
NuojCheng
Collaborator
Loading…
4 tasks done
SFT: tools rendering + developer role in chat-template preprocessing
#4419
opened Jul 10, 2026 by
pekach-d
Loading…
4 tasks done
Refactor MaxText launch scripts and add DeepSeek3 configurations.
#4418
opened Jul 10, 2026 by
copybara-service
Bot
Loading…
docs(nnx): add native LoRA Gemma4 and Qwen3 tutorial notebooks
#4417
opened Jul 10, 2026 by
RexBearIU
Collaborator
Loading…
4 tasks done
test(moe): relax hidden-state gradient atol for ring_ragged_sort
#4415
opened Jul 10, 2026 by
mesakhcienet
Collaborator
Loading…
4 tasks done
Relax tolerances in DiLoCo test to fix TPU7x failure
#4413
opened Jul 10, 2026 by
hsuan-lun-chiang
Collaborator
•
Draft
4 tasks done
[ DeepSeek-V4 ] training FLOPs and MFU calculation
gemini-review
#4410
opened Jul 9, 2026 by
parambole
Collaborator
Loading…
4 tasks done
feat(rl): OOM-safe chunked log-probs for RL actor (enable train_micro>8)
#4409
opened Jul 9, 2026 by
lokic233
Loading…
Avoid Possible MXLA hangs by eagerly scaling up in Elasticity
#4407
opened Jul 9, 2026 by
SujeethJinesh
Collaborator
Loading…
4 tasks done
fix(nnx): optimize scan memory by excluding read-only parameters from…
#4405
opened Jul 9, 2026 by
hsuan-lun-chiang
Collaborator
•
Draft
4 tasks done
fix: resolve vLLM and RL training gibberish outputs for Gemma 3/4 scanned layers
#4404
opened Jul 9, 2026 by
mesakhcienet
Collaborator
•
Draft
4 tasks
Previous Next
ProTip!
no:milestone will show everything without a milestone.