-
Notifications
You must be signed in to change notification settings - Fork 297
Pull requests: radixark/miles
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
feat(offload): support disk target for training-actor offload
#1575
opened Jul 4, 2026 by
Zhichenzzz
Contributor
Loading…
fix(deps): cap transformers<5.13 — 5.13.0 breaks all CPU CI via qwen3_asr collision
#1574
opened Jul 4, 2026 by
guapisolo
Collaborator
Loading…
Add Search-R1 group-relative length penalty
#1573
opened Jul 4, 2026 by
ys-2020
Contributor
Loading…
Add --restore-weights-from-fp32-main: drop the pinned weight backup in colocate
#1572
opened Jul 3, 2026 by
yueming-yuan
Collaborator
•
Draft
GLM-5.2 744B on GB300: sm103 kernel fix, TP8xPP4xDP2xEP16 topology
#1571
opened Jul 3, 2026 by
yueming-yuan
Collaborator
•
Draft
Support externally-triggered checkpoint save via sentinel file
#1570
opened Jul 3, 2026 by
yueming-yuan
Collaborator
Loading…
(6/7) test(session): InloopSessionServer replaces the in-process SessionServer test harness
#1566
opened Jul 3, 2026 by
guapisolo
Collaborator
Loading…
(5/7) feat(session): process-lifecycle layer — supervisor spawns N session workers + router
#1565
opened Jul 3, 2026 by
guapisolo
Collaborator
Loading…
test(session): session-server overhead benchmark under tests/manual/session
#1564
opened Jul 2, 2026 by
guapisolo
Collaborator
Loading…
(3/7) feat(session): strip R3 replay payloads from client chat responses
run-ci-sglang
#1563
opened Jul 2, 2026 by
guapisolo
Collaborator
Loading…
feat(trainer): defer multimodal CUDA transfer for VLM training
#1562
opened Jul 2, 2026 by
TSunny007
Contributor
Loading…
refactor(utils): make loss mask generation strategy-based
#1561
opened Jul 2, 2026 by
TSunny007
Contributor
Loading…
[debug] Add NanScanner: inline NaN/Inf localization utility
#1560
opened Jul 2, 2026 by
yueming-yuan
Collaborator
•
Draft
Add GLM-5/5.1/5.2 (744B MoE) LoRA RL support
run-ci-lora
run-ci-megatron
run-ci-model-scripts
Run model script smoke tests
#1559
opened Jul 1, 2026 by
yushengsu-thu
Collaborator
Loading…
refactor(skills): reframe doc-first-principle as doc-dev (doc↔code consistency)
#1557
opened Jul 1, 2026 by
guapisolo
Collaborator
Loading…
[AMD] build sgl-router-for-miles from source and register test_r3_baseline on CI
#1556
opened Jul 1, 2026 by
JessicaJiang-123
Contributor
Loading…
[fsdp] Gate the true-on-policy log_probs==rollout_log_probs CI assert on ci_disable_logprobs_checker
#1555
opened Jul 1, 2026 by
Zhichenzzz
Contributor
Loading…
[fsdp] PERF: hand bf16 logits to the cross-entropy, drop the full-vocab fp32 buffer
#1554
opened Jul 1, 2026 by
Zhichenzzz
Contributor
Loading…
[fsdp] Actor: re-assert checkpoint over clobbered params via the post_load_fixups registry
#1553
opened Jul 1, 2026 by
Zhichenzzz
Contributor
Loading…
[fsdp] Actor: resolve a PrecisionPolicy and thread its dtypes + fp32-master through apply_fsdp2
#1552
opened Jul 1, 2026 by
Zhichenzzz
Contributor
Loading…
[fsdp] Actor: drive the packing registry at config-lifetime and post-load
#1551
opened Jul 1, 2026 by
Zhichenzzz
Contributor
Loading…
[fsdp] Actor: apply HF-compat class_patches before model construction
#1550
opened Jul 1, 2026 by
Zhichenzzz
Contributor
Loading…
[fsdp] Route FSDP->rollout weight sync through gather_full_param + the WeightBridge transform registry
#1549
opened Jul 1, 2026 by
Zhichenzzz
Contributor
Loading…
Previous Next
ProTip!
Exclude everything labeled
bug with -label:bug.