-
Notifications
You must be signed in to change notification settings - Fork 253
Pull requests: radixark/miles
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[fix] Drain rollout engines before releasing memory on offload
#1335
opened Jun 12, 2026 by
EazyReal
Loading…
Retry transient Ray ActorUnavailableError during rollout engine bringup
#1333
opened Jun 12, 2026 by
EazyReal
Loading…
[fix] Authenticate SGLang engine control-plane and router calls (and fix literal "Bearer None" header)
#1332
opened Jun 12, 2026 by
EazyReal
Loading…
[fix] set deepseek v32 override_hf_native=True
#1330
opened Jun 12, 2026 by
yueming-yuan
Collaborator
Loading…
[refactor] use begin/end_weight_update instead of post_process_weights
run-ci-low-precision
run-ci-megatron
#1329
opened Jun 12, 2026 by
yueming-yuan
Collaborator
Loading…
fix(DrGRPO): replace hardcoded pg-loss divisor with first-class --pg-loss-divisor
#1328
opened Jun 12, 2026 by
EazyReal
Loading…
fix(chat-template): harden tool-call argument decoding against adversarial args
#1327
opened Jun 12, 2026 by
EazyReal
Loading…
fix: make compute_pass_rate ragged-safe at both train and eval call sites
#1326
opened Jun 12, 2026 by
EazyReal
Loading…
fix(rollout): count each rollout once in GRPO group baseline under fan-out
#1325
opened Jun 12, 2026 by
EazyReal
Loading…
fix(rollout): apply --rollout-sample-filter-path generically in the manager
#1324
opened Jun 12, 2026 by
EazyReal
Loading…
[fix] stop merging agentic turns at first non-COMPLETED turn
#1323
opened Jun 12, 2026 by
Shi-Dong
Contributor
Loading…
[OPD] [4/N] Teacher ensembles + exact tail-bucket top-k KL + scoring robustness
#1322
opened Jun 11, 2026 by
maocheng23
Contributor
Loading…
ROCm/support test_deepep_fp8: e2e docs, aiter/sglang patches, mori rollout harness on gfx950
#1320
opened Jun 11, 2026 by
kailashg26
•
Draft
feat: add FlashQLA backend for Qwen GDN linear-attention layers
#1318
opened Jun 11, 2026 by
Zhichenzzz
Contributor
Loading…
fix: load Qwen 3.5 checkpoint with unfused experts
#1317
opened Jun 10, 2026 by
lawrence-harmonic
Contributor
Loading…
[OPD] [3/N] Multi-teacher routing: per-sample teacher selection via --opd-teacher-urls
#1314
opened Jun 9, 2026 by
maocheng23
Contributor
Loading…
fix(qwen3-vl): per-segment mRoPE + vision under CP + THD packing
#1308
opened Jun 8, 2026 by
Zhichenzzz
Contributor
Loading…
fix(mtp): track megatron mtp_model_layer rename in raw converters
#1307
opened Jun 8, 2026 by
Zhichenzzz
Contributor
Loading…
DO NOT MERGE: CI test
run-ci-model-scripts
Run model script smoke tests
#1306
opened Jun 8, 2026 by
yueming-yuan
Collaborator
Loading…
Inject rank and millisecond timestamp into Ray train actor log lines
#1303
opened Jun 7, 2026 by
fzyzcjy
Collaborator
Loading…
Previous Next
ProTip!
Add no:assignee to see everything that’s not assigned.