vllm_dissag: GLM-5.1-FP8 (MLA+DSA) MoRI-EP WideEP disaggregated enablement - #206

Open

MIR-AMD wants to merge 11 commits into

ROCm:developfrom

MIR-AMD:glm5.1-wideep-mad

Commits on Aug 18, 2026

vllm_dissag: GLM-5.1-FP8 (MLA+DSA) MoRI-EP WideEP disaggregated enablement

authored and
MIR-AMD
committed
0b6b349
vllm_dissag: make NIAH harness thinking-model-aware (GLM-5.1)

authored and
MIR-AMD
committed
ed4b16f
vllm_dissag: NIAH multi-seed support (NIAH_SEEDS) for variance-aware accuracy

authored and
MIR-AMD
committed
059d933
vllm_dissag: NIAH gate robust to cold-start JIT (warmup + readiness probe)

authored and
MIR-AMD
committed
a5ea09d
[GLM-5.1] v0.27 level-set: 3.4x decode speedup, 200K context, RDMA + launcher fixes

authored and
MIR-AMD
committed
e0a4b8b
[GLM-5.1] Default WITH_NIXL=0 (lean MoRI-EP-only image)

authored and
MIR-AMD
committed
63fb87d
[GLM-5.1] Add long-context NIAH harness + vllm-disagg operational playbook

authored and
MIR-AMD
committed
af878ed
[GLM-5.1] Remove dead-end MoRI EP env plumbing and its misleading comments

authored and
MIR-AMD
committed
f6dc7af
[GLM-5.1] Decode CUDA-graph FULL_AND_PIECEWISE: 2.1x (EP8) / 1.7x (EP16) faster decode

authored and
MIR-AMD
committed
638a033
[GLM-5.1] Merge hygiene: drop lab leftovers, safe patcher defaults.
MIR-AMD
committed
c1686f5
[GLM-5.1] Point the recipe at the v0.27 image it was validated on.
MIR-AMD
committed
3d166d4