Module hf_cache 

HuggingFace hub cache inspection, repair, and model download.

This module owns all on-disk state management for locally cached HuggingFace models and exposes a progress-callback-based download API that is completely decoupled from any application framework.

1. Call list_local_hf_models / list_supported_hf_models / delete_local_hf_model directly from your command handlers.
2. Call download_model with an Arc<dyn Fn(ModelDownloadProgress)> that forwards progress to whatever event system your app uses.

Structs

LocalHfModel

Information about a locally cached HuggingFace model.

LocalHfModelsResponse

Response returned by list_local_hf_models.

ModelDownloadProgress

Progress payload passed to the download_model callback on every poll tick and on completion.

SupportedHfModel

A supported model that may or may not be downloaded locally.

SupportedHfModelsResponse

Response returned by list_supported_hf_models.

Functions

clean_stale_lock_files

Remove stale .lock and orphaned .part files from the HF cache blobs/ directory for a model.

delete_local_hf_model

Delete a locally cached HuggingFace model by removing its directory from the hub cache.

diagnose_hf_cache

Log the complete HF cache structure for a model to aid debugging.

download_model

ensure_model_cache_dir

Pre-create the HuggingFace cache directory tree for model_id so that the progress-monitor thread can find a valid path immediately, even before hf-hub creates it during the first download.

ensure_snapshot_entries

Ensure snapshot entries exist for known GGUF files.

list_local_hf_models

Scan the local HuggingFace hub cache and return all downloaded models that the inference engine supports.

list_supported_hf_models

Return the full list of models the inference engine supports, together with flags indicating whether each one is fully downloaded, partially downloaded, or not present at all.

model_cache_path

Return the cache sub-directory path for a given model ID.

repair_hf_cache_symlinks

Repair broken symlinks in the HF hub cache for a given model.