Skip to content

Update first token time calculation in cuda benchmark #16324

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
Gasoonjia wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Update first token time calculation in cuda benchmark #16324

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
15061d7
init
Gasoonjia Dec 18, 2025
597c2ce
Merge branch 'main' into solve-zero-first-token-time
Gasoonjia Dec 18, 2025
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
8 changes: 7 additions & 1 deletion .ci/scripts/cuda_benchmark.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,7 +73,13 @@ def parse_pytorch_observer_log(log_line: str) -> Optional[RunMetrics]:
else 0
)
model_load_time_ms = model_load_end_ms - model_load_start_ms
first_token_latency_ms = first_token_ms - prompt_eval_end_ms

# First token latency (TTFT): time from inference start to first token
# For multimodal models (e.g., whisper, voxtral, gemma3), this includes:
# 1. Encoding time (image/audio processing)
# 2. Prefill time (decoder processing encoder outputs)
# This represents the end-to-end user experience of waiting for the first token
first_token_latency_ms = first_token_ms - inference_start_ms

return RunMetrics(
generated_tokens=generated_tokens,
Expand Down
Loading