Skip to content

Add Kimi K2.5 1P1D AIPerf recipe - #257

Draft
esmeetu wants to merge 1 commit into
NVIDIA:mainfrom
esmeetu:kimi-k25-1p1d-aiperf-recipe
Draft

Add Kimi K2.5 1P1D AIPerf recipe#257
esmeetu wants to merge 1 commit into
NVIDIA:mainfrom
esmeetu:kimi-k25-1p1d-aiperf-recipe

Conversation

@esmeetu

@esmeetu esmeetu commented Jul 10, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Add a GB200 vLLM/Dynamo Kimi K2.5 NVFP4 1P1D PDCP8/DDCP8 recipe.
  • Use the public sprint AgentX image inferactinc/public:sprint-agentx-arm64-cu13.0.1-90e0957.
  • Run the AgentX workload through aiperf profile via the upstream custom benchmark runner.

Notes

  • model.path uses the kimi-k2.5-nvfp4 model alias, which should resolve from the local srtslurm.yaml.
  • Upstream does not currently register benchmark.type: aiperf; this recipe uses benchmark.type: custom and invokes aiperf profile directly.

Validation

  • uv run srtctl dry-run -f recipes/vllm/kimi-k2.5/agentx/disagg-gb200-1p1d-pdcp8-ddcp8-c128.yaml

@esmeetu
esmeetu force-pushed the kimi-k25-1p1d-aiperf-recipe branch 2 times, most recently from 9d1f317 to 139568c Compare July 10, 2026 03:28
@esmeetu
esmeetu force-pushed the kimi-k25-1p1d-aiperf-recipe branch from 139568c to d158268 Compare July 10, 2026 03:28
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant