Skip to content

[Klaud Cold] Update glm5.2-fp4-b200-sglang-agentic-mtp SGLang image to nightly-dev-cu13-20260908-20ca564b #2864

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
functionstackx wants to merge 2 commits into
base: main
Choose a base branch
from
+14 −4
Open

[Klaud Cold] Update glm5.2-fp4-b200-sglang-agentic-mtp SGLang image to nightly-dev-cu13-20260908-20ca564b #2864

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
498be85
[Klaud Cold] Update glm5.2-fp4-b200-sglang-agentic-mtp SGLang image t…
functionstackx Sep 7, 2026
57d2ed5
Re-pin to the 2026-09-08 cu13 nightly, which carries sglang#38318 (EA…
functionstackx Sep 8, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 5 additions & 4 deletions configs/nvidia-master.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9007,17 +9007,18 @@ glm5.2-fp4-b300-sglang-agentic-mtp:
# head (num-steps 3, eagle-topk 1, 4 draft tokens = 3 speculative tokens) with
# acceptance pinned to the golden AL 2.99 (golden_al_distribution/glm5.2_mtp.yaml,
# thinking_on, K=3) through SGLANG_SIMULATE_ACC_*. SGLang v0.5.16 is the first
# release that reads SGLANG_SIMULATE_ACC_TOKEN_MODE. The pinned 2026-09-01
# nightly includes FlashInfer 0.6.18's BF16 TRTLLM MoE allocation fix, which is
# required by GLM-5.2's unquantized EAGLE draft head at CUDA-graph capture.
# release that reads SGLANG_SIMULATE_ACC_TOKEN_MODE. The pinned 2026-09-08
# nightly carries forward FlashInfer 0.6.18's BF16 TRTLLM MoE allocation fix
# (first picked up via the 2026-09-01 nightly), which is required by GLM-5.2's
# unquantized EAGLE draft head at CUDA-graph capture.
#
# Same single arm and concurrency grid as glm5.2-fp4-b300-sglang-agentic-mtp:
# cookbook low-latency TP8 with HiCache host-DRAM offload, conc
# [1, 4, 8, 12, 16] (steps of at least 2, hard stop at 16). TP8-only for
# memory as well as comparability -- the ~433 GB NVFP4
# checkpoint needs ~54 GB/GPU across 8 B200s and does not fit below 8.
glm5.2-fp4-b200-sglang-agentic-mtp:
image: lmsysorg/sglang:nightly-dev-cu13-20260901-07c8f729
image: lmsysorg/sglang:nightly-dev-cu13-20260908-20ca564b
model: nvidia/GLM-5.2-NVFP4
model-prefix: glm5.2
runner: cluster:b200-nscale
Expand Down
9 changes: 9 additions & 0 deletions perf-changelog.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6921,3 +6921,12 @@
- "Pick up the latest automatic ROCm DeepSeek-V4 optimizations, including fused mHC post/pre plus RMSNorm, gfx950 C4A top-k dispatch, fused C4 compressor GEMMs, fused SWA q/kv RMSNorm plus q FP8 quantization, and medium-batch cooperative top-k tuning."
- "Keep the existing VLLM_ROCM_USE_AITER=1, VLLM_ROCM_USE_AITER_MOE=1, and --moe-backend aiter settings, and explicitly add VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS=1 plus VLLM_ROCM_QUICK_REDUCE_QUANTIZATION=INT4 to both STP and MTP paths. The current checkpoint's shared-expert path does not satisfy the latest vLLM fusion conditions, so that fusion flag self-disables while preserving recipe parity."
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2792

- config-keys:
- glm5.2-fp4-b200-sglang-agentic-mtp
scenario-type:
- agentic-coding
description:
- "Update SGLang image from lmsysorg/sglang:nightly-dev-cu13-20260901-07c8f729 (2026-09-01 cu13 dev nightly) to lmsysorg/sglang:nightly-dev-cu13-20260907-30705c00 (2026-09-07 cu13 dev nightly, digest sha256:19b8fa1223cc339c1eae7a5b703f1a8c2543b5b119155bf3d7efaef18f77f007, tag commit sgl-project/sglang@30705c00). Docker Hub last pushed the tag at 2026-09-07T01:43:42Z. The newer nightly carries forward the FlashInfer 0.6.18 BF16 TRTLLM MoE allocation fix that the 2026-09-01 pin was taken for. Engine flags, EAGLE MTP settings, golden acceptance length 2.99, HiCache sizing, and the TP8 conc [1, 4, 8, 12, 16] grid are unchanged; the same tag the B200 Qwen3.5 SGLang AgentX recipes moved to in #2861/#2862 and the new FP8 sibling in #2863 starts on."
- "Re-pin from lmsysorg/sglang:nightly-dev-cu13-20260907-30705c00 to lmsysorg/sglang:nightly-dev-cu13-20260908-20ca564b (2026-09-08 cu13 dev nightly, digest sha256:9a352a35c973a2357372e85f3bcb5388b6b3c46c1329165987260f3b089647dc; Docker Hub last pushed 2026-09-08T01:40:59Z, tag commit sgl-project/sglang@20ca564b). The 2026-09-07 build carries an unguarded kv_index_translator.translate_dcp_read_ids call on the DSA fp8 KV read path that the EAGLE draft backend never binds, so GLM-5.2 MTP runs crash intermittently with AttributeError (observed on the MI355X FP8 sibling in run 34173459478 after 74 minutes of serving). sgl-project/sglang#38318 (merged 2026-09-07T20:03Z) adds the None guard and is six commits behind 20ca564b."
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2864
Loading