-
Notifications
You must be signed in to change notification settings - Fork 454
Pull requests: radixark/miles
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Fully-async disaggregated PPO with a full-weight actor: release phase memory, keep the offloaded actor's params, add a 2+2-node example
#3139
opened Sep 7, 2026 by
Michaelsqj
Contributor
Loading…
docs(debug): tell a cold Triton compile from a hang under multi-node colocate
#3132
opened Sep 6, 2026 by
David553
Loading…
test(dist): cover the reloadable process-group destroy/reload cycle
#3131
opened Sep 6, 2026 by
David553
Loading…
fix: isolate trainer disk-offload backups by role
#3130
opened Sep 6, 2026 by
Shi-Dong
Contributor
Loading…
fix: guard connection allocations during offloaded weight sync
#3129
opened Sep 6, 2026 by
Shi-Dong
Contributor
Loading…
fix: use host weights when syncing an offloaded actor
#3128
opened Sep 6, 2026 by
Shi-Dong
Contributor
Loading…
[LoRA] Publish immutable adapter versions alongside generation
run-ci-lora
run-ci-weight-update
Run weight update tests
#3127
opened Sep 6, 2026 by
yushengsu-thu
Collaborator
Loading…
Decide the pipeline stage from the parallel state in compute_advantages_and_returns
#3125
opened Sep 5, 2026 by
Michaelsqj
Contributor
Loading…
Bump sglang to v0.5.19
run-ci-eval
Run eval-labeled CI tests
run-ci-fsdp
run-ci-image
run-ci-lora
run-ci-megatron
run-ci-model-scripts
Run model script smoke tests
run-ci-sglang
run-ci-weight-update
Run weight update tests
#3124
opened Sep 5, 2026 by
Zhichenzzz
Contributor
Loading…
Build model script test arguments from the launcher environment
#3121
opened Sep 5, 2026 by
fzyzcjy
Collaborator
Loading…
Shut the worker manager down on every driver exit path
#3120
opened Sep 5, 2026 by
fzyzcjy
Collaborator
Loading…
Extract the mooncake store config into one table
#3119
opened Sep 5, 2026 by
fzyzcjy
Collaborator
Loading…
Restore the event history before the first process writes events
#3118
opened Sep 5, 2026 by
fzyzcjy
Collaborator
Loading…
Add candidate multi-teacher OPD and a Qwen3.6 short-puzzle example
#3116
opened Sep 5, 2026 by
maocheng23
Contributor
•
Draft
Scan full responses for repetition metrics
#3115
opened Sep 5, 2026 by
Shi-Dong
Contributor
Loading…
Publish the session servers as one list of instance records
#3108
opened Sep 4, 2026 by
nblintao
Contributor
•
1/2
Loading…
DO NOT MERGE: verify Qwen3.5 MTP memory release fix
#3106
opened Sep 4, 2026 by
guapisolo
Collaborator
Loading…
[multi-lora] Push adapter versions without pausing the rollout engines
#3104
opened Sep 4, 2026 by
yushengsu-thu
Collaborator
Loading…
test(docs): guard cli-reference.md against undocumented train flags
#3103
opened Sep 4, 2026 by
bo-lyu
Loading…
4 tasks done
Previous Next
ProTip!
What’s not been updated in a month: updated:<2026-08-07.