Repository navigation
feat(decision): single source of head mount names + head_state_prefixes for the merge strip - #388
Merged
Conversation
…for the merge strip
HEAD_MODULE_NAMES ('task_head', 'scoring_head') becomes the one definition of the
dual mount: attach_transformers / attach_megatron loop over it and reactivate
matches every name instead of hardcoding prefixes.
head_state_prefixes(model) reports which mounts a model actually carries (legacy
single-mount checkpoints -> ('scoring_head',), headless -> ()), unioning Megatron
chunks; merge / export code strips head keys from a merged state dict through it.
state_dict() emits a key under every mounted name (named_parameters deduplicates
shared modules by identity, state_dict does not), so the strip must filter on the
whole tuple, never a single prefix.
tastelikefeet
approved these changes
Nov 17, 2026
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
PR type
PR information
HEAD_MODULE_NAMES ('task_head', 'scoring_head') becomes the one definition of the dual mount: attach_transformers / attach_megatron loop over it and reactivate matches every name instead of hardcoding prefixes.
head_state_prefixes(model) reports which mounts a model actually carries (legacy single-mount checkpoints -> ('scoring_head',), headless -> ()), unioning Megatron chunks; merge / export code strips head keys from a merged state dict through it.
state_dict() emits a key under every mounted name (named_parameters deduplicates shared modules by identity, state_dict does not), so the strip must filter on the whole tuple, never a single prefix.
Experiment results
Paste your experiment result here(if needed).