Skip to content

ScaleRLT: verify current H20 preparation and faster evidence transport - #35

Closed
0z5a wants to merge 1 commit into
ops/scalerlt-iquest-eager-diagnosticfrom
ops/scalerlt-current-state-preflight
Closed

0z5a wants to merge 1 commit into
ops/scalerlt-iquest-eager-diagnosticfrom
ops/scalerlt-current-state-preflight

Conversation

@0z5a

@0z5a 0z5a commented Oct 6, 2026

Copy link
Copy Markdown
Owner

Native RL startup requires verified model inputs, enough private disk space and a writable destination for complete optimizer states. This records the current H20 preflight: 770 frozen source files, seven pinned Thinking-model files and 93.719 GiB free private XFS, with the existing 0z5a environment unchanged. It also adds the editable ScaleRLT paper and current storage/transport tables.

Identical 10,178,626-byte source archive Upload + independent download + verification Relative speed
Direct TLS 24.183 s 1.00×
Existing Mac HTTP proxy, selected per process 6.761 s 3.58×

Each route has one timed attempt. Both server digests and independent download hashes match the original archive. The new own draft destination is verified writable; its documented quota accommodates 150 parts of 512 MiB. Complete native-state transfer, online learning, fresh-worker recovery and reward convergence remain pending.

Validation: all recorded input hashes and artifact manifest match; the preflight exits naturally with code 0, followed by an independent absent-actor, free-lock and empty-GPU check. The earlier shell-path failure is retained. Ruff passes and the standalone paper compiles with the native editor. Stack base: #34; original numerical/data/optimizer contracts are unchanged.

Signed-off-by: 0z5a <dezhen.lu@student.uni-tuebingen.de>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant