Expand description
On-disk model cache for fast model reloading.
Caches quantized model weights + metadata in a binary format (.oxcache)
for faster cold-start loading vs. re-parsing GGUF files.
Format:
Header: "OXCA" (4 bytes) + version u32 + num_entries u64 + metadata_len u32
Metadata: JSON string (hand-serialised, no serde)
Per entry: name_len u32 + name (UTF-8) + quant_type_len u32 + quant_type + data_len u64 + data bytes
Structs§
- Cache
Entry - An entry in the disk cache, representing one named tensor blob.
- Cache
File Info - Information about one cached model file on disk.
- Cache
Manager - Manages multiple cached model files with LRU eviction.
- Disk
Cache - In-memory representation of a
.oxcachefile.
Enums§
- Disk
Cache Error - Errors produced by disk-cache operations.
Constants§
- CACHE_
MAGIC - Magic bytes identifying an Pictor disk cache file.
- CACHE_
VERSION - Current cache format version.