-
Notifications
You must be signed in to change notification settings - Fork 394
Pull requests: radixark/miles
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
fix(megatron): sync fused shared-outer LoRA main grads
#2723
opened Aug 23, 2026 by
zhoutong-hai
Loading…
[8/8] fix(ray): settle trainer admissions without blocking the manager loop
#2719
opened Aug 23, 2026 by
ashtonchew
Loading…
5 of 6 tasks
add DeepSeek-V4-Flash-0731 support and mxfp4->fp8 converter
#2717
opened Aug 23, 2026 by
yueming-yuan
Collaborator
Loading…
fix: skip base-weight update session for LoRA sync
run-ci-lora
run-ci-lora-native
Run native (raw-mode) LoRA plugin e2e tests
#2715
opened Aug 22, 2026 by
yushengsu-thu
Collaborator
Loading…
Bump sglang to v0.5.18
bypass-fastfail
Disable CI fast-fail: run all stages and all tests to surface every failure
run-ci-image
#2714
opened Aug 22, 2026 by
yueming-yuan
Collaborator
Loading…
test: pin update_weight_version abort_all_requests=False regression
#2713
opened Aug 22, 2026 by
yushengsu-thu
Collaborator
•
Draft
test(e2e): verify multi-turn Anthropic sessions
run-ci-sglang
#2711
opened Aug 22, 2026 by
guapisolo
Collaborator
•
2/2
Loading…
fix(rollout): handle invalid tool arguments
#2709
opened Aug 22, 2026 by
MinhaoLi0318
Contributor
Loading…
fix(lora): let a resumed disaggregated run reach the rollout engine
#2708
opened Aug 22, 2026 by
Arist12
Contributor
Loading…
fix: retain colocated LoRA CUDA-IPC exporter storage
#2707
opened Aug 21, 2026 by
zhoutong-hai
Loading…
DeepSeek-V4 on the bumped Megatron, with a selectable training implementation
#2706
opened Aug 21, 2026 by
yueming-yuan
Collaborator
Loading…
feat(ci): turn on the Blackwell-only tests
run-ci-blackwell-only
Scope: run the tests that can only run on Blackwell (nvfp4 / mxfp8)
#2704
opened Aug 21, 2026 by
yueming-yuan
Collaborator
Loading…
feat(ci): dispatch a CI run to a GPU generation
#2703
opened Aug 21, 2026 by
yueming-yuan
Collaborator
Loading…
fix(ci): key the metric-history baseline on the executing stage
#2702
opened Aug 21, 2026 by
yueming-yuan
Collaborator
Loading…
feat(ci): declare supported GPU generations on every CUDA test
#2701
opened Aug 21, 2026 by
yueming-yuan
Collaborator
Loading…
feat(ci): add stage-c-8-gpu-b200 for the Blackwell runner
#2700
opened Aug 21, 2026 by
yueming-yuan
Collaborator
Loading…
fix(ci): unbreak the documented runner setup on a new host
#2699
opened Aug 21, 2026 by
yueming-yuan
Collaborator
Loading…
fix: pad bshd samples to their own length, not the rollout-global max
#2698
opened Aug 21, 2026 by
EazyReal
Loading…
[rollout] Add typed DP-padding postprocess policy
#2694
opened Aug 21, 2026 by
yushengsu-thu
Collaborator
Loading…
feat(megatron): report peak memory for log_probs and actor_train
#2693
opened Aug 21, 2026 by
EazyReal
Loading…
fix(disk-delta): reject noncanonical tensor layouts
#2692
opened Aug 21, 2026 by
nanjiangwill
Contributor
Loading…
Previous Next
ProTip!
Adding no:label will show everything without a label.