Source code for experimental_experiment.xoptim.patterns_ort

from typing import List


[docs] def get_onnxruntime_patterns( verbose: int = 0, ) -> List["PatternOptimization"]: # noqa: F821 """ Returns a default list of optimization patterns for onnxruntime. It is equal to the following list. .. runpython:: :showcode: :rst: from experimental_experiment.xoptim.patterns_api import pattern_table_doc from experimental_experiment.xoptim.patterns_ort import get_onnxruntime_patterns print(pattern_table_doc(get_onnxruntime_patterns(), as_rst=True)) """ from .activation import ( BiasGeluPattern, BiasSoftmaxPattern, FastGeluPattern, GeluOrtPattern, GeluErfPattern, QuickGeluPattern, ) from .activation_grad import SoftmaxGradPattern from .batch_normalization import OrtBatchNormalizationTrainingPattern from .fused_conv import FusedConvPattern from .fused_matmul import ( FusedMatMulDivPattern, FusedMatMulPattern, FusedMatMulx2Pattern, FusedMatMulTransposePattern, ) # from .llm_optim import RotaryEmbeddingPattern # from .gather_grad import GatherGradPattern from .simplified_layer_normalization import ( SimplifiedLayerNormalizationPattern, SkipLayerNormalizationPattern, ) return [ BiasGeluPattern(verbose=verbose), BiasSoftmaxPattern(verbose=verbose), GeluOrtPattern(verbose=verbose), GeluErfPattern(verbose=verbose), FusedConvPattern(verbose=verbose), FastGeluPattern(verbose=verbose), FusedMatMulPattern(verbose=verbose), FusedMatMulx2Pattern(verbose=verbose), FusedMatMulDivPattern(verbose=verbose), FusedMatMulTransposePattern(verbose=verbose), OrtBatchNormalizationTrainingPattern(verbose=verbose), QuickGeluPattern(verbose=verbose), # RotaryEmbeddingPattern(verbose=verbose), # GatherGradPattern(verbose=verbose), SimplifiedLayerNormalizationPattern(verbose=verbose), SkipLayerNormalizationPattern(verbose=verbose), SoftmaxGradPattern(verbose=verbose), ]