Skip to content

[AMD/ROCm] qwen3.5 fp8 mi355x optimization #910

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
seungrokj wants to merge 9 commits into
base: main
Choose a base branch
from
Open

[AMD/ROCm] qwen3.5 fp8 mi355x optimization #910

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
a11cb7a
qwen3.5 fp8 mi355x optimization
seungrokj Mar 13, 2026
a706a84
qwen3.5 fp8 mi355x optimization
seungrokj Mar 13, 2026
d26ce9c
qwen3.5 fp8 mi355x
seungrokj Mar 16, 2026
1bbaf1f
Merge branch 'main' into srok/qwen3.5fp8_mi355x
seungrokj Mar 16, 2026
28030df
qwen3.5 fp8 perf changelog
seungrokj Mar 16, 2026
188efa9
qwen3.5 fp8 perf changelog
seungrokj Mar 16, 2026
fd5c8d6
qwen3.5 fp8 perf changelog
seungrokj Mar 16, 2026
dcf4337
qwen3.5 fp8 perf changelog
seungrokj Mar 16, 2026
3335a18
prevent pythonbytecode
seungrokj Mar 18, 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
2 changes: 1 addition & 1 deletion .github/configs/amd-master.yaml
100644 → 100755
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -228,7 +228,7 @@ qwen3.5-fp8-mi325x-sglang:
- { tp: 8, conc-start: 4, conc-end: 64 }

qwen3.5-fp8-mi355x-sglang:
image: rocm/sgl-dev:v0.5.8.post1-rocm720-mi35x-20260218
image: rocm/sgl-dev:v0.5.9-rocm720-mi35x-20260315
model: Qwen/Qwen3.5-397B-A17B-FP8
model-prefix: qwen3.5
runner: mi355x
Expand Down
19 changes: 16 additions & 3 deletions benchmarks/single_node/qwen3.5_fp8_mi355x.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,24 +19,37 @@ hf download "$MODEL"

SERVER_LOG=/workspace/server.log
PORT=${PORT:-8888}
MEM_FRAC_STATIC=${MEM_FRAC_STATIC:-0.8}
CHUNK_SIZE=8192

# Start GPU monitoring (power, temperature, clocks every second)
start_gpu_monitor

python3 -m sglang.launch_server \
set -x
sglang serve \
--attention-backend triton \
--model-path $MODEL \
--host=0.0.0.0 \
--port $PORT \
--tensor-parallel-size $TP \
--trust-remote-code \
--mem-fraction-static 0.8 > $SERVER_LOG 2>&1 &
--mem-fraction-static $MEM_FRAC_STATIC \
--kv-cache-dtype fp8_e4m3 \
--mamba-ssm-dtype bfloat16 \
--cuda-graph-max-bs $CONC \
--max-running-requests $CONC \
--chunked-prefill-size $CHUNK_SIZE \
--max-prefill-tokens $CHUNK_SIZE \
--disable-radix-cache \
--num-continuous-decode-steps 2 \
--tokenizer-worker-num $TP \
> $SERVER_LOG 2>&1 &

SERVER_PID=$!

# Wait for server to be ready
wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID"

export PYTHONDONTWRITEBYTECODE=1
run_benchmark_serving \
--model "$MODEL" \
--port "$PORT" \
Expand Down
9 changes: 9 additions & 0 deletions perf-changelog.yaml
100644 → 100755
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -961,3 +961,12 @@
description:
- "Extend MiniMax M2.5 FP8 single-node config for H200 with vLLM v0.16.0 (TP8)"
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/869

- config-keys:
- qwen3.5-fp8-mi355x-sglang
description:
- "Improved fp8 performance"
- "Relevant Issue: https://github.com/sgl-project/sglang/issues/19633"
- "Relevant PR: https://github.com/sgl-project/sglang/pull/19826"
- "Image: rocm/sgl-dev:v0.5.9-rocm720-mi35x-20260315"
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/910