Skip to content

NA: opened by mistake#7

Closed
viraatc wants to merge 1 commit into
NVIDIA:mainfrom
viraatc:add-benchx-example
Closed

NA: opened by mistake#7
viraatc wants to merge 1 commit into
NVIDIA:mainfrom
viraatc:add-benchx-example

Conversation

@viraatc
Copy link
Copy Markdown

@viraatc viraatc commented May 15, 2026

<EDIT - Opened by mistake>

benchx is the RWLT-driven variant of slurm_dynamo_trtllm_disagg.yaml:
swaps aiperf for the artificial-analysis Real-World Load Test, runs the
dynamo frontend in approximate-KV router mode (--router-mode kv
--no-router-kv-events --router-ttl-secs 480), wires Eagle3 speculative
decoding into both ctx and gen engine configs, and exposes the HOSTCACHE
and WORKER_METRICS knobs from the bench shell scripts. Default
CONCURRENCY sweep is 1,2,3,6,8,10,16,32,48,64,80,96,112,128,144,160.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@viraatc viraatc closed this May 16, 2026
@viraatc viraatc changed the title Add dynamo TRT-LLM benchx workflow example NA: opened by mistake May 16, 2026
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