Datasetpublic

promotion/aaai27-ronpo-flagship-resume-20260713

AAAI-27 RONPO flagship resume bundle This public dataset repository stores the frozen training/precompute artifacts, protocols, status snapshot, and source snapshot needed to resume the matched Qwen3-8B flagship experiment in a new container. The sealed-test prompts are intentionally not public. Their local-server backup has SHA-256 52b4028bd3ce095524e3ae66f49bf495d1236fea4635248b4263f9db1920df69 and size 436341 bytes; the public split_manifest.json records the split hashes.… See the full description on the dataset page: https://huggingface.co/datasets/promotion/aaai27-ronpo-flagship-resume-20260713.

sourceHugging Faceotherupdated 3mo agoView on Hugging Face

0likes90downloads

Dataset Card

AAAI-27 RONPO flagship resume bundle

This public dataset repository stores the frozen training/precompute artifacts, protocols, status snapshot, and source snapshot needed to resume the matched Qwen3-8B flagship experiment in a new container.

The sealed-test prompts are intentionally not public. Their local-server backup has SHA-256 52b4028bd3ce095524e3ae66f49bf495d1236fea4635248b4263f9db1920df69 and size 436341 bytes; the public split_manifest.json records the split hashes.

Only canonical data-*.arrow datasets are retained. Rebuildable Hugging Face cache files, temporary Arrow files, upstream model caches, local W&B cache, and pool_gate.jsonl are excluded. Model checkpoints are stored separately in the public promotion/qwen3-8b-aaai27-flagship-* model repositories recorded in metadata/protocol/hf_uploads.jsonl.

Download with huggingface-cli download promotion/aaai27-ronpo-flagship-resume-20260713 --repo-type dataset and restore the directory layout under the new flagship experiment root. Reload the upstream Qwen3-8B and ArmoRM models from their public repositories.