Add ray data option for video benchmarks

[oyilmaz-nvidia]

Add ray_data variants for video nightly benchmarks

Summary

Adds ray_data executor coverage for the four video pipelines that currently only run on Xenna in nightly benchmarks, so both backends are tracked side-by-side every night — matching the dual-executor pattern already used for audio_readspeech_*, image_curation_*, domain_classification_*, etc.

Config-only change to benchmarking/nightly-benchmark.yaml — no Python edits needed because video_pipeline_benchmark.py already accepts --executor={xenna,ray_data} and routes through setup_executor().

Changes

Renamed each existing video benchmark to add an explicit _xenna suffix and added a sibling _raydata entry (identical args except --executor=ray_data):

Pipeline	Xenna entry	Ray Data entry	Timeout	num_clips	Min throughput
Embedding	video_embedding_xenna	video_embedding_raydata	400s	1400	4.0/s
Transcoding	video_transcoding_xenna	video_transcoding_raydata	400s	1400	5.0/s
Captioning	video_captioning_xenna	video_captioning_raydata	1800s	377	0.25/s
TransNetV2 + filters	video_transnetv2_motion_aesthetic_filter_embeddings_xenna	video_transnetv2_motion_aesthetic_filter_embeddings_raydata	800s	113	0.25/s

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[greptile-apps]

Greptile Summary

This PR adds ray_data executor variants for the four existing video pipeline benchmarks (embedding, transcoding, captioning, transnetv2_motion_aesthetic_filter_embeddings), matching the dual-executor pattern used for audio, image, and domain classification pipelines. It is a config-only change — no Python edits.

• Each existing video_* benchmark is renamed with an _xenna suffix, and a sibling _raydata entry is added with --executor=ray_data as the only argument difference, keeping all other flags, timeouts, and requirements identical.
• All four raydata entries correctly inherit the same exact_value clip counts, min_value throughput requirements, and sink_data Slack notifications as their xenna counterparts.

Confidence Score: 5/5

Config-only addition that faithfully mirrors existing xenna entries; no logic changes.

All four raydata entries are verified to be exact copies of their xenna counterparts with only --executor=ray_data substituted. GPU resource flags, argument sets, timeouts, clip count requirements, and throughput thresholds are all consistent across every pair.

No files require special attention.

Important Files Changed

Filename	Overview
benchmarking/nightly-benchmark.yaml	Adds four ray_data benchmark entries mirroring existing xenna entries; renames four existing entries with _xenna suffix. All args, timeouts, requirements, and GPU resource flags are consistent between pairs.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[nightly-benchmark.yaml] --> B[video_pipeline_benchmark.py]
    B --> C{--executor}
    C -->|xenna| D[video_embedding_xenna]
    C -->|ray_data| E[video_embedding_raydata]
    C -->|xenna| F[video_transcoding_xenna]
    C -->|ray_data| G[video_transcoding_raydata]
    C -->|xenna| H[video_captioning_xenna]
    C -->|ray_data| I[video_captioning_raydata]
    C -->|xenna| J[video_transnetv2_..._xenna]
    C -->|ray_data| K[video_transnetv2_..._raydata]

Loading

_{Reviews (1): Last reviewed commit: "Add ray data option for video benchmarks" | Re-trigger Greptile}

[ayushdg]

The throughput might be different for ray data. Do we want to run verify and then add these or add these first and then fix?