nerfbaselines.backends

class nerfbaselines.backends.ApptainerBackend(spec: ApptainerBackendSpec)

Bases: RemoteProcessRPCBackend

install()

name = 'apptainer'

shell(args=None)

class nerfbaselines.backends.ApptainerBackendSpec

Bases: TypedDict

conda_spec: CondaBackendSpec | None

environment_name: Required[str]

home_path: str

image: str | None

python_path: str

class nerfbaselines.backends.Backend

Bases: object

install()

instance_call(instance: int, method: str, *args, **kwargs)

instance_del(instance: int)

name = 'unknown'

shell(args: Tuple[str, ...] | None = None)

static_call(function: str, *args, **kwargs)

class nerfbaselines.backends.CondaBackend(spec: CondaBackendSpec)

Bases: RemoteProcessRPCBackend

install()

name = 'conda'

shell(args=None)

class nerfbaselines.backends.CondaBackendSpec

Bases: TypedDict

environment_name: Required[str]

install_script: str | None

python_version: str | None

class nerfbaselines.backends.DockerBackend(spec: DockerBackendSpec)

Bases: RemoteProcessRPCBackend

install()

name = 'docker'

shell(args=None)

class nerfbaselines.backends.DockerBackendSpec

Bases: TypedDict

build_script: str

conda_spec: CondaBackendSpec

environment_name: Required[str]

home_path: str

image: str | None

python_path: str

replace_user: bool

class nerfbaselines.backends.SimpleBackend

Bases: Backend

instance_call(instance: int, method: str, *args, **kwargs)

instance_del(instance: int)

name = 'python'

static_call(function: str, *args, **kwargs)

nerfbaselines.backends.backend_allocate(size: int) → Tuple[int, memoryview] | None

Allocates a memory block in the shared memory block valid for the next call to the backend. The function is only valid on the worker side handling request. Everywhere else it will return None.

Parameters:

size – The size of the memory block to allocate.

Returns:

If the memory block

was allocated, the function returns a tuple of id and memory view. Otherwise, it returns None.

Return type:

Optional[Tuple[int, memoryview]]

nerfbaselines.backends.backend_allocate_ndarray(shape, dtype)

nerfbaselines.backends.get_backend(method_spec: MethodSpec, backend: str | None) → Backend

nerfbaselines.backends.get_implemented_backends(method_spec: MethodSpec) → Sequence[Literal['conda', 'docker', 'apptainer', 'python']]

nerfbaselines.backends.get_mounts()

nerfbaselines.backends.mount(ps: str | Path, pd: str | Path)

nerfbaselines.backends.run_on_host()

nerfbaselines.backends.zero_copy(zero_copy: bool = True)

A context manager that enables zero-copy for the current thread. Zero-copy is used for all subsequent calls for all backends. A zero-copy mode instructs the backend to reuse the shared memory used for data transfer. This is useful when the data is large and speed is important. However, it is important to only use results of backend calls inside the context manager. The data will be overwritten by subsequent calls.

Parameters:

zero_copy – If True, zero-copy is enabled for the current thread. If False, zero-copy is disabled.