onnx_diagnostic.helpers.model_builder_helper

onnx_diagnostic.helpers.model_builder_helper.create_model_builder(config: Any, model: torch.nn.Module, cache_dir: str, precision: str = 'fp32', execution_provider: str = 'cpu', verbose: int = 0, **extra_options) → Model

Creates a model based on a configuration. The onnx model is returned by function save_model_builder().

Parameters:

• config – configuration

• cache_dir – cache directory

• precision – precision

• execution_provider – execution provider

• verbose – verbosity

• extra_options – extra options

Returns:

model

onnx_diagnostic.helpers.model_builder_helper.download_model_builder_to_cache(url: str = 'https://raw.githubusercontent.com/microsoft/onnxruntime-genai/refs/heads/main/src/python/py/models/builder.py')

Downloads builder.py from the https://github.com/microsoft/onnxruntime-genai/blob/main/src/python/py/models/builder.py.

onnx_diagnostic.helpers.model_builder_helper.find_names_pattern(names: List[str]) → str

Finds a repeatable patterns in a list of names. It tries to locate the figures.

<<<

from onnx_diagnostic.helpers.model_builder_helper import find_names_pattern

pattern = find_names_pattern(["past_key_values_key_0", "past_key_values_key_1"])
print(pattern)

>>>

    past_key_values_key_%d

onnx_diagnostic.helpers.model_builder_helper.import_model_builder(module_name: str = 'builder') → object

Imports the downloaded model.by.

onnx_diagnostic.helpers.model_builder_helper.make_genai_config(config, onnx_filename: str) → Dict

Creates genai config file for a model.

Parameters:

• config – configuration from transformers

• onnx_filename – onnx configuration

Returns:

configuration

onnx_diagnostic.helpers.model_builder_helper.save_model_builder(self, out_dir: str | None = '', verbose: int = 0) → str | ModelProto

Saves a model created by function create_model_builder(). If out_dir is empty or not specified, the function still returns the generated model.