Verlog hiring-environment checkpoint archive
Original checkpoint files, retaining their run/step directory structure. This is an experiment archive, not a single ready-to-load Transformers model.
archive_manifest.json lists every archived file with its size and SHA-256.
All archived files were checked against Hub metadata, and two small files were
downloaded and checked. This is not a distributed training-resume test.
Some original checkpoints contain only model weights or critic probes; others include optimizer, RNG, dataloader, and replay/source state. Archiving preserves what existed but does not make incomplete checkpoints resumable. Restore the whole relevant run/step folder and use its matching training code/configuration.
PyTorch checkpoint files may use pickle; only load files from trusted sources.