[PD] Re-add early _engine_paused gate before decode_queue / pop_boots… - #24908

Closed

Zhichenzzz wants to merge 42 commits into

sgl-project:sglang-milesfrom

Zhichenzzz:zhichen/pd-pause-early-gate

Commits on Apr 5, 2026

[diffusion] CI: improve diffusion comparison benchmark setting for realistic perf and auto-discover ut (#22086)

authored and
Fridge003
committed
43654ef
[Fix] Fix nightly tests (#22140)
Fridge003
committed
164bc0a
Enable IndexCache for DeepSeek V3.2 (#21405)

jinyouzhi
and
Fridge003
committed
07f57fc
fix: TRT-LLM MHA CUDA illegal address with EAGLE v2 + DP attention (#21649)

committed
c1927e1
[Hotfix] Fix router gemm on sm103 (#22134)

Fridge003
authored and
Kangyan-Zhou
committed
1519acf

Commits on Apr 6, 2026

[1/8] [sglang-miles] True on-policy training support for FSDP2 (#18639)
yueming-yuan
committed
74c9eab
[2/8] [sglang-miles] R3 (Rollout Routing Replay) DeepEP and MTP support (#18642)
yueming-yuan
committed
453eb15
[3/8] [sglang-miles] Support INT4 QAT for RL (#18565)
yueming-yuan
committed
2bc3f68
[4/8] [sglang-miles] PD disaggregation for RL (#18646)
yueming-yuan
committed
277147c
[5/8] [sglang-miles] MTP related fix (#18647)
yueming-yuan
committed
7af1f15
[6/8] [sglang-miles] tmp fix for vlm training: use legacy_load_mm_data for qwen-vl (#18781)
yueming-yuan
committed
24def86
[7/8] [sglang-miles] support better token id return for TITO (#19731)

guapisolo
authored and
yueming-yuan
committed
69018ba
[8/8] [feat] Support cross turn token in after last user message (#20066)

guapisolo
authored and
yueming-yuan
committed
3606aec
[sglang-miles] fix weight checker (#21494)
yueming-yuan
committed
98b5440
P2P Weight Update features for miles (#21278)

JD-ETH
authored and
yueming-yuan
committed
629aa25
Fix is_multimodal_gen attr not in v0.5.10 ModelConfig
yueming-yuan
committed
d335128
Fix flashinfer fused_moe with topk>8 (#22201)

Qiaolin-Yu
authored and
yueming-yuan
committed
0e3e23e

Commits on Apr 13, 2026

[Hotfix] final fixes for P2P Transfer (#22663)

authored
2b93c2f

Commits on Apr 14, 2026

[fix] Fix deadlock and TOCTOU race in pause-aware weight update locking (#22754)

maocheng23
and
claude
authored
8f874d1

Commits on Apr 16, 2026

[sglang-miles] chore: add bias for base layer with lora (#22846)
gongyisheng
authored
3c43f43
[fix]sglang-miles fully async deadlock (#22623)

maocheng23
and
claude
authored
586421e
R3 Support on PD Disaggregation (mini_lb) (#22916)
zyzshishui
authored
a5bfbcb

Commits on Apr 17, 2026

[miles] Pick up two dp rank fixes from main (#23101)

ByronHsu
and
Byron Hsu
authored
146cf20

Commits on Apr 21, 2026

Commits on Apr 23, 2026

feat: enable SGLANG_PATCH_TOKENIZER by default (#23409)

ByronHsu
authored and
root
committed
9955e47
[minor] Make DEFAULT_FORCE_STREAM_INTERVAL configurable via SGLANG_FORCE_STREAM_INTERVAL (#23215)

ByronHsu
authored and
root
committed
508ef15
[PD+DP] Allow PrefillDelayer in disaggregated-prefill mode (#23588)

ByronHsu
authored and
root
committed
e29c347

Commits on Apr 26, 2026

Fix Qwen3 MoE: also guard EP all-reduce with not use_reduce_scatter (follow-up to #23731) (#23734)

authored and
root
committed
e0790b5

Commits on Apr 27, 2026

Commits on Apr 28, 2026

[sglang-miles] Improve PD pause (#23887)

authored
187ed07
[sglang-miles] Fix PD+R3+Overlap (#23888)

authored
5a8e9bc
[perf] cherry-pick #22911: support return_routed_experts with overlap scheduling (#23860)

authored
bb9223d

Commits on May 9, 2026

Commits on May 10, 2026