-
Notifications
You must be signed in to change notification settings - Fork 7
Pull requests: radixark/miles_diffusion
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[cosmos3] Bitwise train/rollout parity: FSDP precision spec + cosmos3_bitwise rollout patch group
#109
opened Aug 8, 2026 by
zhihengy
Collaborator
Loading…
2 tasks
feat(qwen-image): bitwise train<->rollout parity via autocast-parity norms and a pinned flash kernel
#108
opened Aug 8, 2026 by
zhihengy
Collaborator
Loading…
2 of 3 tasks
refactor(fsdp)(5/5): compile per-model precision plans onto the dtype patch
run-ci-fsdp
Run FSDP domain tests on this PR
#106
opened Aug 8, 2026 by
Rockdu
Collaborator
Loading…
5 of 6 tasks
refactor(diffusion): move rollout patch group selection to a CLI arg
#97
opened Aug 6, 2026 by
Rockdu
Collaborator
Loading…
4 of 6 tasks
fix(nft): fix deterministic mode and add the SD3.5 DiffusionNFT e2e
#87
opened Aug 1, 2026 by
Rockdu
Collaborator
Loading…
6 tasks done
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!
Updated in the last three days: updated:>2026-08-05.