Buckets:
| # Drive extraction to completion. Re-runs in passes so clips become available as | |
| # their broadcast finishes downloading; each pass shards across NSHARD processes | |
| # to saturate the GPU (one process tops out ~155fps, the 3090 does ~380fps). | |
| cd /workspace/scripts_7/scripts | |
| NSHARD=4 | |
| for pass in $(seq 1 40); do | |
| done_n=$(ls /workspace/scripts_7/data/fs_spot/processed/*.npz 2>/dev/null | wc -l) | |
| vids=$(ls /workspace/scripts_7/data/fs_spot/videos/*.webm 2>/dev/null | wc -l) | |
| echo "=== pass $pass | extracted ${done_n}/371 | videos ${vids}/11 ===" | |
| [ "$done_n" -ge 371 ] && { echo "ALL 371 EXTRACTED"; break; } | |
| pids=() | |
| for i in $(seq 0 $((NSHARD-1))); do | |
| python3 -m temporal_scripts.fs_spot_extract --shard $i/$NSHARD > /tmp/shard_$i.log 2>&1 & | |
| pids+=($!) | |
| done | |
| for p in "${pids[@]}"; do wait $p; done | |
| new_n=$(ls /workspace/scripts_7/data/fs_spot/processed/*.npz 2>/dev/null | wc -l) | |
| echo " pass $pass added $((new_n-done_n)) clips -> ${new_n}/371" | |
| # nothing new and downloads still running -> wait for more video to land | |
| if [ "$new_n" -eq "$done_n" ] && [ "$vids" -lt 11 ]; then echo " waiting for downloads..."; sleep 120; fi | |
| done | |
| echo "FINAL: $(ls /workspace/scripts_7/data/fs_spot/processed/*.npz 2>/dev/null | wc -l)/371" | |
Xet Storage Details
- Size:
- 1.27 kB
- Xet hash:
- bf4de6acfbabafcfa353c93617c79c55c8a8b85bfb02edb4e9cb2dca1b030d8a
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.