latency-sensitive-bench/vla-finetuning-scope-data

VLA fine-tuning scope

This archive retains evaluation records, committed paper inputs, exact W&B provenance and historical code snapshots for seven QwenOFT runs. The three formal scope-comparison policies are identified in run_index.json.

Shared training-data owner

Training uses latency-sensitive-bench/latency-transfer-data@5305590625c28a6922e73f46d18d52738762c906, prefix flappy_fix_latency_2_200ep. That shared source remains available to other experiments. This archive references the same data and does not retain another large copy.

The temporary copied Parquet files passed row, episode, image and action-statistics validation before their redundant destination objects were removed. Their source Git/LFS identities remain in the verification receipt. Original train/validation files are still available at the fixed shared revision.

See verification/native_loader_verification.json, verification/data_download_verification.json and provenance/wandb_training_provenance.json.

Experiment repository consolidation

Independent experiment sources are retained under training/<source-repository>/. Each source has its own dataset config and original metadata. Different latency units, episode budgets and action layouts are preserved. Evaluation JSON files are under evaluations/<source-model-repository>/. experiment_repository_index.json records immutable source identities and model/data/paper relationships.

Shared Flappy raw data now have one owner: latency-sensitive-bench/latency-transfer-data@5305590625c28a6922e73f46d18d52738762c906, prefix training/flappy_200ep/. Original repository names in provenance identify sources that were consolidated.