fix(importer): emit all shards for multi-part GGUF models - #9513
Merged
Conversation
The llama-cpp HuggingFace importer iterated files one at a time and kept overwriting `lastGGUFFile`, so sharded repos such as `unsloth/Kimi-K2.6-GGUF` (14 `Q8_K_XL` parts) produced a gallery entry pointing only at the final shard — useless to llama.cpp's split loader, which needs shard 1 to discover the set. Group shards up front via new helpers in `pkg/huggingface-api` (`SplitShardSuffix`, `ShardGroup`, `GroupShards`). The llama-cpp importer now picks a group (preferred quant, then last-group fallback) and emits every shard, with `Model:` pointing at shard 1. `FindPreferredModelFile` returns shard 1 of the first matching group so the gallery agent's preview stays coherent for sharded repos. Adds unit coverage for the HuggingFace branch of the importer (which had none), plus shard-detection tests in the hfapi package. Assisted-by: Claude:Opus-4.7 [Read] [Edit] [Bash]
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 llama-cpp HuggingFace importer iterated files one at a time and kept overwriting
lastGGUFFile, so sharded repos such asunsloth/Kimi-K2.6-GGUF(14Q8_K_XLparts) produced a gallery entry pointing only at the final shard — useless to llama.cpp's split loader, which needs shard 1 to discover the set.Group shards up front via new helpers in
pkg/huggingface-api(SplitShardSuffix,ShardGroup,GroupShards). The llama-cpp importer now picks a group (preferred quant, then last-group fallback) and emits every shard, withModel:pointing at shard 1.FindPreferredModelFilereturns shard 1 of the first matching group so the gallery agent's preview stays coherent for sharded repos.Adds unit coverage for the HuggingFace branch of the importer (which had none), plus shard-detection tests in the hfapi package.
Assisted-by: Claude:Opus-4.7 [Read] [Edit] [Bash]