Modules
Production machine-learning inference for the BEAM.
The MLServe OTP application.
Guarded invocation of MLServe.Model callbacks.
Serves an ordinary Elixir function as a model.
Always returns the same configured result, whatever the input.
Raised when a backend callback raises, throws, or exits.
Coalesces concurrent single predictions into one backend batch call.
Loads the models declared in application configuration when MLServe starts.
Optional TTL cache for inference results.
Reads and validates MLServe configuration.
The prediction hot path: route, admit, cache, hook, dispatch, record.
The structured error MLServe raises and reports.
The behaviour every MLServe model backend implements.
The supervision subtree for one loaded model version.
The model catalog: which models exist, at which versions, and where traffic goes.
Owns one model version's lifecycle: loading, readiness, status and draining.
The normalised, validated description of one loaded model version.
DynamicSupervisor holding one MLServe.ModelInstance per loaded {name, version}.
The slim record read from ETS on every prediction.
Model artifact validation: path containment, size limits, and integrity checks.
MLServe's top-level supervision tree.
Telemetry events emitted by MLServe.
Logs MLServe telemetry events, with no dependencies beyond Logger.
Telemetry.Metrics definitions for MLServe, ready for LiveDashboard or any reporter.
A single inference worker holding backend state.
Supervises one model version's worker pool.