rlmesh.numpy

The NumPy backend: the same clients, models, and sandbox sessions as the top-level package, with tensor leaves decoded to NumPy arrays. Space wrappers from these clients sample NumPy-compatible values.

pip install "rlmesh[numpy]"

Conversion semantics

• asarray() returns a writable copy, matching Gymnasium, where observations are writable (obs /= 255.0 works). For a zero-copy read-only view, use numpy.from_dlpack(tensor) or the buffer protocol.
• from_array() always copies: it makes the array C-contiguous and serializes it into a fresh RLMesh tensor.

Classes

• EnvFactoryNumPy-backed EnvFactory -- the default, named for symmetry.
• ModelNumPy-backed model: predict works in NumPy values.
• RemoteEnvNumPy-backed remote client for a single RLMesh environment.
• RemoteModelNumPy-backed handle to a served model (policy).
• RemoteVectorEnvNumPy-backed remote client for a vectorized RLMesh environment.
• SandboxBuildExperimentalBuild-from-source infrastructure for a sandbox image.
• SandboxEnvExperimentalOwned NumPy-backed sandbox session for one environment.
• SandboxInfoExperimentalInformation about a running RLMesh sandbox container.
• SandboxModelExperimentalA model served from an isolated container.
• SandboxRuntimeExperimentalContainer run-time settings for a sandbox -- docker run flags.
• SandboxVectorEnvExperimentalOwned NumPy-backed sandbox session for vectorized environments.

Functions

• asarray()Return a writable NumPy array copied from an RLMesh tensor.
• ensure_available()Raise if NumPy is not installed.
• from_array()Encode a NumPy array or scalar as an RLMesh value.
• space_from_spec()Create a NumPy-adapted space wrapper for a native space spec.

Type aliases

• NumpyValuePrimitive, NumPy array, or nested RLMesh-compatible value.