Expand description
Identifying a model’s format and capabilities from its files. The full runtime-bid auction lives with the runtime adapters; this module holds the file-format detection those decisions build on.
Re-exports§
pub use format::GgufArchitectureProfile;pub use format::GgufFacts;pub use format::ModelFormat;pub use format::gguf_architecture_profile;pub use format::ollama_chat_profile;pub use format::ollama_profile;pub use format::ollama_vision_profile;pub use gguf::gguf_facts;pub use gguf::gguf_general_architecture;pub use gguf::has_ggml_magic;pub use gguf::has_gguf_magic;pub use identification_cache::IdentificationCache;pub use identity::IdentifiedModel;pub use identity::RuntimeBid;pub use identity::identify;pub use pipelines::DiffusersPipelineProfile;pub use pipelines::PipelineFamily;pub use pipelines::PipelineFamilyRegistry;pub use pipelines::PipelineRefinement;pub use pipelines::SchedulerFacts;pub use safetensors::safetensors_format;pub use safetensors::safetensors_header_format;
Modules§
- format
- Format and capability facts derived from a model’s files.
- gguf
- A minimal GGUF header reader: enough to pull the architecture, context length, chat-template presence, and file type without loading the weights.
- identification_
cache - A cache over
identifykeyed by an on-disk freshness signature. Identifying a file model reads its GGUF/safetensors header and stats its directory; over a large shelf that repeats on every resolution pass. The cache skips it when a model’s bytes are unchanged. - identity
- The identity and bid foundation types: what
Identification::identifyproduces about a model (IdentifiedModel) and what a runtime adapter offers to serve it (RuntimeBid). Theidentifyorchestration and the bid auction build on these. - pipelines
- The diffusers pipeline-family registry: maps a
model_index.json_class_nameto a modality/capability/parameter profile, refined by the model’s scheduler and repo name. This is how a discovered diffusers bundle gets its image/video/audio shape and its sampler parameter schema. - safetensors
- Reading a safetensors header to tell an MLX-format weight directory from a plain one.