Skip to content

[WIP] Updated DSv4 vllm B300 MTP #1271

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
wzhao18 wants to merge 3 commits into
base: main
Choose a base branch
from
+24 −2
Draft

[WIP] Updated DSv4 vllm B300 MTP #1271

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
99753d6
add DP to b300 mtp
wzhao18 May 4, 2026
e5c062d
Update changelog
wzhao18 May 4, 2026
09a7294
Update Docker image version for dsv4-fp4-b300-vllm-mtp
wzhao18 May 4, 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: 5 additions & 1 deletion .github/configs/nvidia-master.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2730,7 +2730,7 @@ dsv4-fp4-b300-trt:
- { tp: 8, ep: 8, dp-attn: true, conc-start: 256, conc-end: 1024 }

dsv4-fp4-b300-vllm-mtp:
image: vllm/vllm-openai:v0.20.0-cu130
image: vllm/vllm-openai:v0.20.1
model: deepseek-ai/DeepSeek-V4-Pro
model-prefix: dsv4
runner: b300
Expand All @@ -2744,12 +2744,16 @@ dsv4-fp4-b300-vllm-mtp:
search-space:
- { tp: 4, conc-start: 1, conc-end: 256, spec-decoding: mtp }
- { tp: 8, conc-start: 1, conc-end: 64, spec-decoding: mtp }
- { tp: 4, ep: 4, dp-attn: true, conc-start: 256, conc-end: 2048, spec-decoding: mtp }
- { tp: 8, ep: 8, dp-attn: true, conc-start: 4096, conc-end: 8192, spec-decoding: mtp }
- isl: 8192
osl: 1024
search-space:
- { tp: 4, conc-start: 1, conc-end: 64, spec-decoding: mtp }
- { tp: 8, conc-start: 1, conc-end: 64, spec-decoding: mtp }
- { tp: 4, ep: 4, conc-start: 64, conc-end: 256, spec-decoding: mtp }
- { tp: 4, ep: 4, dp-attn: true, conc-start: 256, conc-end: 512, spec-decoding: mtp }
- { tp: 8, ep: 8, dp-attn: true, conc-start: 1024, conc-end: 2048, spec-decoding: mtp }

qwen3.5-fp8-h200-sglang:
image: lmsysorg/sglang:v0.5.9-cu129-amd64
Expand Down
13 changes: 12 additions & 1 deletion benchmarks/single_node/dsv4_fp4_b300_vllm_mtp.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,17 @@ if [ "${EP_SIZE:-1}" -gt 1 ]; then
EP_ARGS=(--enable-expert-parallel)
fi

MAX_NUM_BATCHED_TOKENS=$(( ISL * 2 ))
MOE_ARGS=()
if [ "${DP_ATTENTION}" = "true" ]; then
MOE_ARGS=(--moe-backend deep_gemm_mega_moe)
fi

if [ "${DP_ATTENTION}" = "true" ]; then
MAX_NUM_BATCHED_TOKENS=2048
else
MAX_NUM_BATCHED_TOKENS=$(( ISL * 2 ))
fi

BENCHMARK_MAX_MODEL_LEN=$MAX_MODEL_LEN

if [ "${EVAL_ONLY}" = "true" ]; then
Expand All @@ -61,6 +71,7 @@ vllm serve "$MODEL" --host 0.0.0.0 --port "$PORT" \
--block-size 256 \
--no-enable-prefix-caching \
"${EP_ARGS[@]}" \
"${MOE_ARGS[@]}" \
--compilation-config '{"cudagraph_mode":"FULL_AND_PIECEWISE","custom_ops":["all"]}' \
--attention_config.use_fp4_indexer_cache True \
--tokenizer-mode deepseek_v4 \
Expand Down
7 changes: 7 additions & 0 deletions perf-changelog.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2179,3 +2179,10 @@
- "Use vllm/vllm-openai:v0.20.1-ubuntu2404 directly for GB200 MTP2 instead of upgrading vLLM inside the v0.20.0 container"
- "Fix applies to all 7 multinode launch scripts, the benchmark-multinode-tmpl workflow, and process_result.py"
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1256

- config-keys:
- dsv4-fp4-b300-vllm-mtp
description:
- "Update image tag to vllm/vllm-openai:v0.20.1-cu130"
- "Add DEP configs for B300 vLLM MTP"
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1271
Loading