latency-sensitive-bench/Standard-Pipeline

Standard Pipeline

Environment adaptation, training evidence, latency profiles, raw demonstrations and evaluation traces, organized by environment and experiment stage.

• —AirRaid: zero-latency and profile-latency experiments.

Only raw demonstrations are distributed. Generate converted training datasets on the training server. Model weights remain in the dedicated model repository and are referenced from the experiment records. Pin a commit revision for reproducible downloads.

• —inverted_pendulum

• —ant

• —hopper

• —half_cheetah

• —walker2d

• —humanoid

All seven environments use condition/model-family/experiment/stage directories, with each environment’s shared zero-latency data stored once. Existing dataset configuration names and split names are retained and point to the reorganized raw files. Recorded source clocks and terminated experiments remain explicitly documented. This organization is the default main archive layout.

Archive cleanup

Cleanup ledger · Current archive inventory. The pre-cleanup archive remains available at revision db063e5a24a56aff85ed2f1c79dac8f77ad05d57. Dataset configuration names, split names and raw data paths are unchanged by this cleanup.

This archival view retains latencies.jsonl and removes its byte-identical infer_latencies.jsonl alias. Runtime logging documentation describes runtime output, where both names can still be emitted. End-to-end timing logs are retained separately. Historical checksum manifests and source scripts describe their original revisions; use the current inventory for this cleaned branch.