You need to agree to share your Hugging Face information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Latest remaining-experiment handoff — 1 November 2026

Use OPERATOR_INSTRUCTIONS_20261101.txt and B300-remaining-20261101-r2-data-ready.tar.gz for the next wave. Verify its .sha256 before extraction. The20261030instructions below describe the historical run and do not govern this wave.

NIER_IPMF_NEW only. CIFAR10 seed1 ONLY, MLP384x10,Brownian reference,20cycles x9000updates/direction. No ASBM training jobs. Seven TabPC datasets and Rossmann use remaining seeds1–4.33jobs configured for verified data/B300batch calibration;20new TabDiT requests remain blocked on original data/schema. Full matrix: EXPERIMENTS_20261101.csv.

CIFAR's verified original data archive, receipt and download instructions are uploaded. The r2-data-ready downloader uses this HF copy by default. Complete its checksum verification before paid training. Source and queue changes passed CPU smoke/dry runs; B300timing and full50kFID remain unmeasured. The8GPUqueue co-locates eligible jobs and refills automatically. No4hourcompletion guarantee.

Download the corrected release:

hf download Peer007/NIER-reproducibility-b300 B300-remaining-20261101-r2-data-ready.tar.gz B300-remaining-20261101-r2-data-ready.tar.gz.sha256 OPERATOR_INSTRUCTIONS_20261101.txt --repo-type dataset --local-dir handoff_20261101
cd handoff_20261101
shasum -a 256 -c B300-remaining-20261101-r2-data-ready.tar.gz.sha256
tar -xzf B300-remaining-20261101-r2-data-ready.tar.gz
cd B300-remaining-20261101
cat OPERATOR_INSTRUCTIONS.txt

Keep all previous results and source archives. CIFAR protocol,campaign report,prepublication HF audit.

Historical dataset card follows.

B300 replacement campaign — revision 3

After dataset preparation and preflight, run one command from this directory:

bash START_B300.sh --inputs INPUTS.json --clients-per-gpu 8 --max-workers 32

The detached queue launches in strict priority order, shares GPUs between small experiments, automatically replaces finished/failed workers, and stops within four hours. No agent is required. No production training has been launched here.

Read OPERATOR_INSTRUCTIONS.txt for preparation, the complete order, monitoring, capacity overrides and full-results upload. Use --dry-run to inspect all 43 jobs. SCHEDULER_R3_CHANGES.txt describes validation and limits. Scientific budgets and models are unchanged. Old r1/r2 archives do not contain this launcher.

Downloads last month: 886

Total file size:

744 GB