Skip to content

Turned on the v2 fused top-k for GLM-5.x on ROCm #2915

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

Draft
EricKing626 wants to merge 2 commits into
base: main
Choose a base branch
from
Draft

Turned on the v2 fused top-k for GLM-5.x on ROCm #2915

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
999cc4d
Turned on the v2 fused top-k for GLM-5.x on ROCm
EricKing626 Sep 9, 2026
0f79a36
Add perf-changelog entry for GLM-5.2 FP4 MI355X top-k v2
EricKing626 Sep 10, 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
6 changes: 2 additions & 4 deletions benchmarks/single_node/agentic/glm5.2_fp4_mi355x_sglang_mtp.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,10 +62,8 @@ export AIPERF_HTTP_TCP_USER_TIMEOUT=900000
# Outlast the client pool so the race cannot occur.
export SGLANG_TIMEOUT_KEEP_ALIVE=900
# The DSA indexer's top-k v2 kernel (default since v0.5.14) is JIT-compiled
# from CUDA-only source (cooperative_groups.h) and cannot build for gfx950;
# v1 dispatches to the precompiled HIP op in sgl-kernel (upstream MI355X CI
# runs DSA models the same way).
export SGLANG_OPT_USE_TOPK_V2=false
# PR #36684 & PR #36851 turned the v2 fused top-k on for GLM-5.x on ROCm
export SGLANG_OPT_USE_TOPK_V2=true

# HiCache L2 (host DRAM), optionally extended with Mooncake L3.
# KV_OFFLOADING=dram requires KV_OFFLOAD_BACKEND=hicache or mooncake.
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 @@ -6969,3 +6969,12 @@
- "Use SGLang's current W4A4 MegaMoE and DP LM-head flags for DP-attention instead of the deprecated MegaMoE environment variables."
- "Resolve draft-model jobs through the existing B200 SGLang speculative recipe and the staged DeepSeek-V4-Pro-0813 checkpoint."
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2821

- config-keys:
- glm5.2-fp4-mi355x-sglang-agentic-mtp
scenario-type:
- agentic-coding
description:
- "Set SGLANG_OPT_USE_TOPK_V2=true in benchmarks/single_node/agentic/glm5.2_fp4_mi355x_sglang_mtp.sh. The recipe forced the v1 top-k because the DSA indexer's v2 fused kernel would not build for gfx950; sgl-project/sglang PR #36684 and PR #36851 add the ROCm path, so v2 now dispatches natively."
- "No other recipe change: image, TP4/EP4 HiCache and TP8/EP1 resident arms, MTP, and concurrency grids are unchanged."
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2915