Skip to content

Fix B200 Dynamo vLLM recipe concurrencies #1305

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

Merged
Oseltamivir merged 2 commits into from
May 10, 2026
Merged

Fix B200 Dynamo vLLM recipe concurrencies #1305

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
5371f4a
fix: sync B200 Dynamo vLLM recipe concurrencies
Oseltamivir May 9, 2026
aa35e78
conc recipe
Oseltamivir May 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: 3 additions & 3 deletions .github/configs/nvidia-master.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7958,7 +7958,7 @@ dsv4-fp4-b200-dynamo-vllm:
search-space:
# B200 adaptation of the DSV4 GB200 vLLM disagg recipes. Each worker
# maps to one full 8-GPU B200 node.
- conc-list: [1, 64, 128]
- conc-list: [1, 16, 32, 64, 128]
prefill:
num-worker: 1
tp: 8
Expand All @@ -7971,7 +7971,7 @@ dsv4-fp4-b200-dynamo-vllm:
tp: 8
ep: 1
dp-attn: false
- conc-list: [1024, 2048, 4096, 8192]
- conc-list: [256, 512, 1024]
prefill:
num-worker: 2
tp: 8
Expand All @@ -7984,7 +7984,7 @@ dsv4-fp4-b200-dynamo-vllm:
tp: 8
ep: 8
dp-attn: true
- conc-list: [8192]
- conc-list: [8192, 12345]
prefill:
num-worker: 3
tp: 8
Expand Down
2 changes: 1 addition & 1 deletion ...arks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/8k1k/disagg-b200-high-tpt-megamoe.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -122,7 +122,7 @@ benchmark:
type: "sa-bench"
isl: 8192
osl: 1024
concurrencies: "4096"
concurrencies: "256x512x1024"
req_rate: "inf"
use_chat_template: true
custom_tokenizer: "sa_bench_tokenizers.vllm_deepseek_v4.VLLMDeepseekV4Tokenizer"
Expand Down
2 changes: 1 addition & 1 deletion benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/8k1k/disagg-b200-low-latency.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -124,7 +124,7 @@ benchmark:
type: "sa-bench"
isl: 8192
osl: 1024
concurrencies: "1"
concurrencies: "1x16x32x64x128"
req_rate: "inf"
use_chat_template: true
custom_tokenizer: "sa_bench_tokenizers.vllm_deepseek_v4.VLLMDeepseekV4Tokenizer"
Expand Down
2 changes: 1 addition & 1 deletion ...marks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/8k1k/disagg-b200-max-tpt-megamoe.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -122,7 +122,7 @@ benchmark:
type: "sa-bench"
isl: 8192
osl: 1024
concurrencies: "4096"
concurrencies: "8192x12345"
req_rate: "inf"
use_chat_template: true
custom_tokenizer: "sa_bench_tokenizers.vllm_deepseek_v4.VLLMDeepseekV4Tokenizer"
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 @@ -2323,3 +2323,10 @@
- "Turn off explicit deep_gemm_mega_moe backend selection in B300 Dynamo vLLM throughput recipes"
- "Let vLLM choose the MoE backend automatically to avoid CUDA symmetric-memory rendezvous failures during startup"
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1304

- config-keys:
- dsv4-fp4-b200-dynamo-vllm
description:
- "Fix B200 Dynamo vLLM recipe benchmark concurrencies to match the nvidia-master.yaml search space"
- "Propagate low-latency concurrencies 1/64/128, high-throughput concurrencies 1024/2048/4096/8192, and max-throughput concurrency 8192 into the srt-slurm recipe files"
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1305
Loading