-
Notifications
You must be signed in to change notification settings - Fork 8
Pull requests: radixark/miles_diffusion
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
perf(rollout): six independent rollout-path fixes and knobs
#198
opened Aug 21, 2026 by
Rockdu
Collaborator
Loading…
3 of 6 tasks
feat(fsdp): add --fsdp-reshard-after-forward for ZeRO-2 style training
#196
opened Aug 21, 2026 by
Rockdu
Collaborator
Loading…
3 of 6 tasks
[cosmos3] Bitwise train/rollout parity: FSDP precision spec + cosmos3_bitwise rollout patch group
run-ci-e2e
Run e2e metric-regression tests on this PR
#109
opened Aug 8, 2026 by
zhihengy
Collaborator
Loading…
2 tasks
fix(fsdp): serialize weight-sync buckets only on the gather src to stop CUDA-IPC pinning
#65
opened Jul 25, 2026 by
zhihengy
Collaborator
Loading…
fix(rollout): make engine port range start configurable via MILES_ROLLOUT_START_PORT
#64
opened Jul 24, 2026 by
zhihengy
Collaborator
Loading…
fsdp: assert weight-sync grouping covers the train world
#36
opened Jul 13, 2026 by
zhihengy
Collaborator
Loading…
ProTip!
Follow long discussions with comments:>50.