convert: keep the model-level hc head mixer in --mtp exports - #1
Open
drluoto wants to merge 1 commit into
Open
Conversation
The mtp_only whitelist kept only embed/norm/lm_head, but llama_model_qwen4exp requires the model-level output_hc_norm/_down/_up at load, so a detached sidecar exported by --mtp failed with 'tensor output_hc_norm.weight not found'. Keep model.hyper_connection_mixer.* as well (34 -> 37 tensors); verified loading and generating on gfx1151/ROCm with both a self-converted sidecar and the pre-existing community one. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
1 task
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The
mtp_onlywhitelist kept only embed/norm/lm_head, butllama_model_qwen4exprequires the model-leveloutput_hc_norm/_down/_upat load, so a detached sidecar exported by--mtpfails withtensor 'output_hc_norm.weight' not found.This keeps
model.hyper_connection_mixer.*as well (34 → 37 tensors). Verified end-to-end on gfx1151/ROCm together with crusaderky's detached-head loader fix: sidecar loads, greedy output clean at 2.7k-token prompts, measured numbers in the upstream PR thread.(Same fix I posted as a comment there — sending it as a proper PR so it can ride along when you update ggml-org#27836.)