Repository navigation
validator: a testnet single GPU mode where the probe agent shares the card and may be swapped for a quantised build - #170
Merged
Merged
Conversation
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Context compression required a second GPU for the probe agent, which a one card testnet validator cannot give it. New flags, also read from the environment:
--judge-share-gpu(MT_JUDGE_SHARE_GPU=1) runs the probe agent on the same card at half its memory;--judge-modeland--judge-revision(MT_JUDGE_MODEL, MT_JUDGE_REVISION) swap the probe agent for another repository, refused on any network but test;--slm-memory-share(MT_SLM_MEMORY_SHARE) caps the model under test. The container entrypoint demands gVisor only when the sandbox runtime is runsc, so MT_SANDBOX_RUNTIME=runc works on testnet. Documented in docs/validator_setup.md.🤖 Generated with Claude Code