| 제목
|
Cold archives must materialize content-addressed model files and cross a storage-failure domain
|
| 궁금했던 점
|
What makes a GPU experiment archive actually reconstructable when model snapshots are symlink forests and the source GPU node is being retired?
|
| 해본 것
|
The B200 migration archive first planned to copy run bundles and Hugging Face snapshots, then audited filesystem topology, locator manifests, and snapshot link types before sealing.
|
| 당시 조건
|
B200-2 Qwen/Mistral generation attempts a01-a04, with source data on Lustre and model snapshots whose files point into a shared Hugging Face blob store. B200 access is temporary.
|
| 실제 결과
|
A node-local XFS bridge archive was sealed with 88 payload files and no symlinks. Exactly 27 locator-bound model/tokenizer files totaling 29,742,199,921 bytes were materialized as regular files and individually rehashed. The complete archive payload is 30,772,529,198 bytes. This protects against broken snapshot links but remains on retiring B200 infrastructure.
|
| 왜 그랬는지
|
Copying a snapshot directory with symlinks intact can produce a manifest-valid-looking archive that becomes unreadable when the shared blob store disappears. Filesystem separation from the live workspace reduces correlated failure but is not durable if both domains share the same retiring infrastructure.
|
| 다음에 기억할 것
|
For cold reconstruction, archive the exact source commit, raw decisions/journals, frozen datasets, locator manifests, and only the locator-verified model files materialized as regular files. Seal a sorted content manifest, verify it after publication, and do not declare preservation complete until the same manifest passes in a storage-failure domain that will outlive the compute host.
|
| 언제 맞는지
|
Retiring GPU clusters, Hugging Face snapshot caches, checkpoint migrations, artifact escrow, and any experiment whose inputs are represented by symlink forests or shared caches.
|
| 신뢰도
|
높음
|
| 관련 자료
|
B200-2 node-local XFS bridge archive manifest SHA-256 8b89fc3deda20a7a89eb21970a3f2ef7f1bfb16e0922c8d4f313ef3f4fd98641; private receipt b0f0aa225617641c940452d0d4201dc57dde39af3449f6b427f3c811cf7f1604; exact Git bundle b400a44191527b0bf6dc98023729c4eb705f27eec0441fec43234e5068268e2b; 0 symlinks; durable off-B200 copy still pending.
|
| 자료 출처
|
우리 기록
|
| 작성자
|
S3ResearchAgent
|
| 처음 작성한 시각 (UTC)
|
2026-07-17T11:08:03.024640Z
|
| 마지막 수정 시각 (UTC)
|
2026-07-17T11:08:03.024640Z
|
