AGILLM 512B native checkpoint step 5024

Owner-authorized PUBLIC archival backup, 2026-11-01. Native int8 latent magnitudes, table, checkpoint/optimizer/RNG/state, exact source, launcher and hotmods are preserved. Saved at13:41:48UTC after loss guard. This is NOT a quality endorsement or authorization to restart; preserve HOLD.

Completion and integrity

Until ARCHIVE_COMPLETE.json exists this backup is INCOMPLETE. manifest.json maps independently zlib-compressed chunks to original relative /workspace paths with compressed and raw SHA256, offsets and file hashes. All remote chunks must match pinned Hub object SHA256/size; small objects are downloaded and SHA256 checked, large objects also receive anonymous range-read verification. This is not a full remote restore or training-resume test.

Restore

Use archived chunk_archive.py restore with manifest and downloaded chunks, into a NEW empty destination. Restore preserves original workspace-relative paths; relocate carefully because exact source/lineage pins and absolute paths matter. Keep .HOLD/HALT.json; do not automatically start supervisor. Build custom kernels from included source with matching runtime versions.

Scope/exclusions

No credentials, raw dataset contents/cache, private personal history, live environment files, unrelated checkpoints or transient process locks. Dataset identifiers, contracts, progress and tokenizer are included; exact data replay additionally requires the referenced datasets/files at their expected layout. Prior inference packs, logs and profiling outputs omitted; optional snapshot/dollpack export history is not a native resume dependency. Existing loss regression and ineffective automatic-brake wiring require separate diagnosis before restart. Original bytes are not modified.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support