Skip to content

[AMD][Qwen3.5] Use HiCache kernel IO and page_first on MI355X MXFP4 AgentX #2972

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
yichiche wants to merge 5 commits into
base: main
Choose a base branch
from
+14 −4
Open

[AMD][Qwen3.5] Use HiCache kernel IO and page_first on MI355X MXFP4 AgentX #2972

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
3d9f4d5
[AMD][Qwen3.5] Pin MI355X MXFP4 AgentX to sglang-rocm 20260910
yichiche Sep 10, 2026
a744721
[AMD][Qwen3.5] Fix AgentX launch on sglang-rocm 0.5.19: cuda-graph-ma…
yichiche Sep 11, 2026
1ce26e1
[AMD][Qwen3.5] AgentX INT4 prefill allreduce and recv-interval 60
yichiche Sep 11, 2026
ef7191b
[AMD][Qwen3.5] Bump MI355X MXFP4 AgentX image to sglang-rocm 20260911
yichiche Sep 11, 2026
2fa9a88
[AMD][Qwen3.5] Collapse AgentX changelog into one short #2972 entry
yichiche Sep 12, 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: 3 additions & 3 deletions benchmarks/single_node/agentic/qwen3.5_fp4_mi355x_sglang_mtp.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@ check_env_vars \
MODEL TP CONC EP_SIZE KV_OFFLOADING \
TOTAL_CPU_DRAM_GB RESULT_DIR DURATION

SCHEDULER_RECV_INTERVAL=${SCHEDULER_RECV_INTERVAL:-30}
SCHEDULER_RECV_INTERVAL=${SCHEDULER_RECV_INTERVAL:-60}

if [[ -n "${SLURM_JOB_ID:-}" ]]; then
echo "JOB $SLURM_JOB_ID running on ${SLURMD_NODENAME:-unknown}"
Expand Down Expand Up @@ -90,7 +90,7 @@ export SGLANG_USE_AITER=1
export SGLANG_USE_AITER_UNIFIED_ATTN=1
export AITER_FLYDSL_FORCE=1
export SGLANG_MAMBA_SSM_DTYPE=bfloat16
export ROCM_QUICK_REDUCE_QUANTIZATION=INT8
export ROCM_QUICK_REDUCE_QUANTIZATION=INT4
export SGLANG_TIMEOUT_KEEP_ALIVE=1800

if [ "${EVAL_ONLY:-false}" != "true" ]; then
Expand All @@ -113,7 +113,7 @@ SGLANG_CMD=(
--watchdog-timeout 1200
--page-size 16
--kv-cache-dtype fp8_e4m3
--cuda-graph-max-bs "$CUDA_GRAPH_MAX_BS"
--cuda-graph-max-bs-decode "$CUDA_GRAPH_MAX_BS"
--max-running-requests "$MAX_RUNNING_REQUESTS"
--max-prefill-tokens 16384
--chunked-prefill-size 16384
Expand Down
2 changes: 1 addition & 1 deletion configs/amd-master.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -361,7 +361,7 @@ qwen3.5-fp4-mi355x-sglang-mtp:
- { tp: 4, conc-start: 4, conc-end: 16, spec-decoding: mtp }

qwen3.5-fp4-mi355x-sglang-agentic-mtp:
image: lmsysorg/sglang-rocm:v0.5.19-rocm720-mi35x-20260908
image: lmsysorg/sglang-rocm:v0.5.19-rocm720-mi35x-20260911
model: amd/Qwen3.5-397B-A17B-MXFP4
model-prefix: qwen3.5
runner: cluster:mi355x-amds
Expand Down
10 changes: 10 additions & 0 deletions perf-changelog.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7408,3 +7408,13 @@
- "Enable required power for the seven existing B200 Kimi-K3 recipes, use the pinned AgentX producer and shared collector, fix the exporter import URI, and record DCP8 with offload disabled to match the serving commands."
- "启用 Kimi-K3 B200 实测功耗并修正 DCP 与 Offload 元数据。"
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/3043

- config-keys:
- qwen3.5-fp4-mi355x-sglang-agentic-mtp
scenario-type:
- agentic-coding
description:
- "Bump image to lmsysorg/sglang-rocm:v0.5.19-rocm720-mi35x-20260911."
- "Rename --cuda-graph-max-bs to --cuda-graph-max-bs-decode for sglang 0.5.19."
- "Prefill allreduce INT4; scheduler-recv-interval default 60."
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2972