sima_utils.transformer.devkit.utils

Attributes

VLM_EVAL_PRINT_INOUTS
VLM_EVAL_PROFILE
mla_rt_api

Classes

MLAGpio
MLABuffer	A wrapper around the Mla_buffer defined in devkit apis with additional buffer properties.
MLAModel	Model definition with buffers for ifms and ofms.

Functions

connect_mla_rt_api(lib_mla_rt_path)
disconnect_mla_rt_api()

Module Contents

sima_utils.transformer.devkit.utils.VLM_EVAL_PRINT_INOUTS

sima_utils.transformer.devkit.utils.VLM_EVAL_PROFILE

sima_utils.transformer.devkit.utils.mla_rt_api: mla_rt_service.devkit_python_api.Mla_rt_apis | None = None

sima_utils.transformer.devkit.utils.connect_mla_rt_api(lib_mla_rt_path: pathlib.Path | None)

sima_utils.transformer.devkit.utils.disconnect_mla_rt_api()

class sima_utils.transformer.devkit.utils.MLAGpio(lib_mla_rt_path: pathlib.Path)

lib_mla_rt_path

mla_poke_path

mla_control_gpio(pin: int, value: int) → None

mla_read_reg(reg_addr: str) → int

mla_write_reg(reg_addr: str, value: str) → None

class sima_utils.transformer.devkit.utils.MLABuffer

A wrapper around the Mla_buffer defined in devkit apis with additional buffer properties.

name: str

shape: tuple[int, Ellipsis]

dtype: str

align_last_dim: bool = True

mla_buffer: mla_rt_service.devkit_python_api.Mla_buffer | None = None

property itemsize: int

Returns the size of the buffer item in bytes.

property size: int

Returns the size of the buffer in bytes.

allocate()

Creates and allocates the buffer.

free()

Frees the buffer if it is allocated.

upload(data: numpy.ndarray)

Uploads the data to the buffer.

download(begin: tuple[int, Ellipsis] | None = None, shape: tuple[int, Ellipsis] | None = None) → numpy.ndarray

Downloads the data from the buffer.

get_buf_addr_len(begin: tuple[int, Ellipsis] | None = None, shape: tuple[int, Ellipsis] | None = None) → mla_rt_service.devkit_python_api.DADDR_LEN

Creates a DADDR_LEN object for the slice of the buffer defined by begin and shape.

class sima_utils.transformer.devkit.utils.MLAModel

Model definition with buffers for ifms and ofms.

model_path: pathlib.Path

ifms: list[MLABuffer | tuple[MLABuffer, tuple[int, Ellipsis] | None, tuple[int, Ellipsis] | None]]

ofms: list[MLABuffer | tuple[MLABuffer, tuple[int, Ellipsis] | None, tuple[int, Ellipsis] | None]]

ifm_buffer_addr_lens: list[mla_rt_service.devkit_python_api.DADDR_LEN | None] = []

ofm_buffer_addr_lens: list[mla_rt_service.devkit_python_api.DADDR_LEN] = []

unique_models: ClassVar[dict[pathlib.Path, Any | None]]

load() → Any

Loads the model and caches the model object so that the model is only loaded once.

free()

Frees the model if it is loaded.

run(update_ifm_addr_len_dict: dict[int, Any] | None = None) → float

Loads and runs the model. Updates the ifm addr_len if needed. Returns the runtime of the model.

classmethod load_all_models(do_parallel_load: bool = False, elf_dir: pathlib.Path | None = None)