serving: attest image split from the runtime; per-card attestation pays N cards per hotkey; sampled audit - #1727
Merged
Merged
Conversation
…ys N cards per hotkey; sampled audit - docker/attest is its own image (entrius/gt-attest, attest-image.yml): one long-running container per box beside the runtime, on every GPU; runtime images no longer carry the sidecar. GET /info self-report (telemetry only). - attest.py judges every card the sidecar reports: capacity = passing cards, score = speed credit x cards, UUID dedupe across all cards; a card without the model resident (free VRAM above total - 0.8 x reservation) does not count. - Verification is sampled per (hotkey, round): every baseline prompt and failed request, plus max(10, 20%) of the completed gateway requests. - Loadout carries runtime_image (digest, written by the conformance job) and attest.image; both persisted per round for the /compute release card. SERVING_ATTEST_REFERENCE_URL + compose `attest` service for the reference. - Conformance on Lium rents a second pod for the attest image; check_serving_runtime --attest-url. Claude-Session: https://claude.ai/code/session_01FJtQK6NCjUgYkGTopczEf4
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
docker/attestis its own image (entrius/gt-attest:<tag>, newattest-image.yml): one container per box beside the runtime, on every GPU. The sparkinfer image no longer ships the sidecar.GET /infoself-report (telemetry only, nothing pays on it).attest.pyjudges every card the sidecar reports:capacity= passing cards,score = speed credit × cards, UUID dedupe across all cards. New check: the model must be resident on the card (free VRAM ≤ total − 0.8 × reservation,SERVING_ATTEST_MODEL_RESIDENT_RATIO) — a bare second GPU earns nothing.SERVING_AUDIT_SAMPLE_MIN/_FRACTION).runtime_image(digest, written by the conformance job) andattest.image, both persisted per round (needs gittensor-dbserving/attest-image-columns, dasserving/attest-image, uicompute/attest-image, docsdocs/attest-image-multi-card).SERVING_ATTEST_REFERENCE_URL, composeattestservice (GT_ATTEST_TAG). Lium conformance rents a second pod for the attest image;check_serving_runtime --attest-url. Miner:SERVING_BACKEND_CONCURRENCYenv override for multi-card boxes.To land: dispatch
attest-image.ymlwith tagv1, re-run sparkinfer conformance soruntime_imagegets its digest, testnet miners switch to two containers.https://claude.ai/code/session_01FJtQK6NCjUgYkGTopczEf4