fix(models): detect nvfp4 experts behind a mixed-precision compressed-tensors format - #343
Merged
Conversation
cadocruz
pushed a commit
to cadocruz/FreeToken
that referenced
this pull request
Sep 2, 2026
Traz 3 commits oficiais novos: - feat(glm5_next): support GLM-5.3-Flash (FlashML-org#332) - fix(hf): download the shards the safetensors index names (FlashML-org#336) - fix(models): detect nvfp4 experts behind a mixed-precision compressed-tensors format (FlashML-org#343) Sem conflito com o trabalho local do fork (sm_120 Blackwell optimizations, 2e79d96) -- confirmado via 'git merge-tree --write-tree', arquivos tocados nao se sobrepoem.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
RedHatAI/GLM-5.3-Flash-NVFP4 now ships quantization_config.format: mixed-precision (the 2026-08-28 re-export added an fp8 MTP layer); nvfp4-pack-quantized only appears on the routed-expert config group. detect_expert_quant only read the top-level field, so expert_quant became compressed-tensors and ft serve failed with no expert-bank provider for expert_quant='compressed-tensors'.
Fix: for mixed-precision, take the format from the config group that targets the experts (all groups only when none does).