Repository navigation
Conversation
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
sachinkaushik
previously approved these changes
Sep 22, 2026
…d mark The orchestrator always passed --explicit-end-mark, so the harness told kiosk-core when each turn ended. Endpointing delay is part of voice-to-voice latency and part of what a hands-free customer waits through, so forcing it out of the measurement made the published number structurally unable to show it. Natural endpointing is now the default. --explicit_end_mark is still available for isolating processing latency, but it is opt-in and documented as excluding endpointing. DEFAULT_RUNS drops from 12 to 8 so that the default sweep over all scripted conversations lands on a turn count that supports a p95. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Supports voice-enabled-interactions PR intel-retail#112 review item 13. kiosk-core now stamps every turn with one stable stream id instead of the per-turn session_id, so consolidated_metrics.csv collapses from one "Based on 1 total calls" row per turn against an opaque UUID into the single aggregate row the consolidator exists to produce. That breaks this script's own collector, which stored one timestamp per id and would have reported a single transaction for the whole run, so pair events as a stack instead: sort by timestamp and match each end to the most recent unmatched start. This also still handles the older per-turn ids. kiosk-core emits for every completed turn, including the discarded warmup -- production instrumentation has no concept of a warmup. This script already filtered those out of its own summary via since_ms, but consolidate-metrics reads the logs directly and was averaging warmup turns in. Prune them once the stack is down (they cannot be rewritten while kiosk-core holds them open through a RotatingFileHandler), keeping the unfiltered original as *.txt.with-warmup -- deliberately not .txt, so the consolidator's glob does not pick it back up. Verified on an 88-turn run: 99 paired transactions before pruning, 88 after, consolidating to one aggregate row. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Supports voice-enabled-interactions PR intel-retail#112 review item 18. The metrics-collector samples CPU, GPU, NPU and memory for the whole time the stack is up -- image pulls, the health wait, the discarded warmup, teardown -- so an average over the raw counter files describes a mostly-idle stack rather than the turns being measured. Write results/measured_window.json with the epoch-ms span the measured turns ran over, which the application's scripts/trim_metrics_to_window.py uses to narrow those files before consolidating. Best-effort: failing to write it leaves the counters un-trimmed, which is the previous behaviour rather than a failed run. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
3 of 9 tasks
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
PR Checklist
What are you changing?
Issue this PR will close
close: #issue_number
Anything the reviewer should know when reviewing this PR?
Test Instructions if applicable
If the there are associated PRs in other repositories, please link them here (i.e. intel-retail/performance-tools )