Repository navigation
Keep dry reads off source weights - #7
Open
GroveMinting wants to merge 1 commit into
Open
GroveMinting wants to merge 1 commit into
GroveMinting wants to merge 1 commit into
Conversation
This was referenced Sep 22, 2026
10xCrazyHorse
added a commit
to yaaa-cool/mini-AGI
that referenced
this pull request
Oct 3, 2026
A read without --save opened the paged pool writable, so dirty experts evicted from a full RAM cache were written into the source weights directory while the trunk, routers, optimiser and manifest were not - a hybrid checkpoint. minagi/dryread.py redirects the pool's expert writes to a temporary directory beside the weights; written experts are read back from there, everything else from the untouched source. cmd_read installs it with a three-line hook when --save is absent. Approach follows upstream PR volotat#7 (sparse overlay) with fewer upstream lines touched. Test: tests/test_bug6_dry_read.py (CPU).
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fixes #6.
What
readwithout--save.Why
A dry read still trains in memory. Previously, when working-set churn overflowed
ram_cache, dirty experts were written directly intoweights/even though the trunk, routers, optimizer bundle, and manifest were not saved. That could leave a hybrid checkpoint. Simply opening the pool read-only is not sufficient because an evicted expert would lose its in-session update.Verification
Four CPU tests pass. They cover:
train.py readcommand against an existing model;Also verified with
compileallandgit diff --check.