API reference¶
Auto-generated from the live Pydantic models and class docstrings via
mkdocstrings. Source of truth is the
code in src/sheaf/; this section never drifts from a release because
the docs build imports the package itself.
| Module | What's there |
|---|---|
| Specs | ModelSpec, ResourceConfig |
| API contracts | Per-model-type request / response schemas, AnyRequest, AnyResponse, BaseRequest, BaseResponse, ModelType |
| Scheduling | BatchPolicy, bucket_requests |
| Caching | CacheConfig, ResponseCache |
| LoRA | LoRAAdapter, LoRAConfig, parse_source, resolve_active_adapters |
| Batch runner | BatchSpec, BatchRunner, JsonlSource, JsonlSink |
| Worker | WorkerSpec, SheafWorker, JobQueueClient, JobQueue, ResultStore, RedisStreamsQueue, RedisHashResultStore |
| Client | SheafClient, AsyncSheafClient, RetryConfig, SheafError |
What's not auto-doc'd¶
Backends in sheaf.backends.* aren't auto-rendered here — the public
contract is the typed request / response shape, not the backend's
internals. The Models catalogue lists every
backend's backend= name, extras flag, and pointer to the runnable
examples/quickstart_*.py. That's the right entry point for users.
For backend implementers (writing your own), see
CONTRIBUTING.md
— the abstract base lives at sheaf.backends.base.ModelBackend.