MCPcopy Create free account

hub / github.com/pytorch/executorch / functions

Functions28,493 in github.com/pytorch/executorch

↓ 21 callersMethodcall_operator
( self, op: EdgeOpOverload, args: tuple[Argument, ...], kwargs: Dict[str, Argu
backends/cadence/aot/decompose_ops.py:32
↓ 21 callersMethodclear
Derestrict AllowedDtypeSet to all allowed dtypes in yaml.
exir/dialects/edge/_ops.py:57
↓ 21 callersMethodgen_blob_data
Generates a buffer with special first and last bytes, repeating the middle byte of the pattern.
exir/_serialize/test/test_program.py:546
↓ 21 callersFunctionget_mapped_key
Map a state dict key using a mapping dictionary with "{}" layer number placeholders.
examples/models/checkpoint.py:118
↓ 21 callersFunctionis_param_node
(exp_prog: ExportedProgram, node: torch.fx.Node)
backends/xnnpack/utils/utils.py:99
↓ 21 callersFunctionparse_etrecord
Parses an `ETRecord` file and returns an `ETRecord` object that contains the deserialized graph modules, program buffer, and a debug handle m
devtools/etrecord/_etrecord.py:744
↓ 20 callersMethodOperators
(self, j)
backends/nxp/backend/ir/lib/tflite/SubGraph.py:123
↓ 20 callersMethod__init__
(self)
backends/apple/metal/tests/test_modules.py:700
↓ 20 callersFunction_compress
Compress (B, H, S, D) tensor to nibble-packed uint8 + fp32 norms.
backends/cuda/tests/test_tq4_sdpa.py:55
↓ 20 callersFunction_count_ops
(gm, target)
backends/mlx/test/test_passes.py:61
↓ 20 callersFunction_make_symint
Create a symbolic dimension backed by the provided ShapeEnv.
backends/arm/test/misc/tosa_dialect/test_tosa_shape_ops.py:21
↓ 20 callersFunction_test_cv_model
( model: torch.nn.Module, test_runner, dtype: torch.dtype, use_dynamic_shapes: bool, )
backends/test/suite/models/test_torchvision.py:32
↓ 20 callersMethod_test_op_activation_case
( self, module, edge_op, inputs, quantize=False, activation=No
backends/xnnpack/test/passes/test_activation_fusion.py:44
↓ 20 callersMethodcall_scalar
Return a scalar value for the current pass stage. In transform-for-annotation passes this returns the Python scalar directly. In late
backends/arm/_passes/arm_pass.py:167
↓ 20 callersMethodcheck_common_constraints
Checks common xnnpack constraints Args: node (torch.fx.Node): Node to check common constraints against ep (E
backends/xnnpack/partition/config/xnnpack_config.py:122
↓ 20 callersFunctiondeserialize_pte_binary
Returns a PTEFile deserialized from the given runtime binary data.
exir/_serialize/_program.py:747
↓ 20 callersMethodflush
(self)
examples/mediatek/aot_utils/llm_utils/loggingprinter.py:29
↓ 20 callersMethodget_freqs
Get the precomputed frequencies for the given input position and sequence length. Args: input_pos (torch.Tensor): The in
examples/models/llama/rope.py:353
↓ 20 callersMethodmatches
Check if this step fully matches the given node.
backends/mlx/pattern_utils.py:156
↓ 20 callersFunctionmodule_factory
(function: Callable)
backends/arm/test/models/test_nn_functional.py:31
↓ 20 callersMethodnamed_parameters
(self)
extension/training/pybindings/_training_module.py:67
↓ 20 callersMethodnormalize
Normalize a given string using the `EnglishTextNormalizer` class. Which preforms commons transformation on english text.
examples/mediatek/aot_utils/llm_utils/tokenizers_/tokenization_whisper.py:549
↓ 20 callersMethodrun_memory_planning
( self, original: GraphModule, opt_level: int = 2, mem_algo: int = 1, # greed
backends/cadence/aot/tests/test_memory_passes.py:255
↓ 20 callersFunctionselect_by_module_path
Match nodes whose `nn_module_stack` (the chain of nn.Module attribute names walked to reach this op during tracing) contains a path matching
devtools/intermediate_output_tap/_selectors.py:126
↓ 20 callersMethodslot_map
(self, tree)
backends/mlx/builder/program_builder.py:200
↓ 19 callersMethod__init__
(self)
exir/tests/test_passes.py:1453
↓ 19 callersMethod__init__
(self, verbose: bool = False)
examples/qualcomm/oss_scripts/llama/static_llm_quant_recipe.py:267
↓ 19 callersFunction_is_annotated
Given a list of nodes (that represents an operator pattern), return True if any of the node is annotated, otherwise return False
backends/samsung/quantizer/annotator.py:57
↓ 19 callersMethod_test_cat
(self, module, inputs, cat_num=1, quant=False, quant_ops=2)
backends/xnnpack/test/ops/test_cat.py:29
↓ 19 callersMethod_update_params_qdtype
(self, node: torch.fx.Node, params: Dict)
backends/samsung/builders/node_visitor.py:79
↓ 19 callersFunctionbuild_executorch_binary
( model, inputs, file_name, dataset, quant_dtype: Optional[Precision] = None, skip_op_
examples/mediatek/aot_utils/oss_utils/utils.py:23
↓ 19 callersFunctionconstant_prop_pass
This pass is for constant propagation for Exported Program with lifted parameters, as the parameters will not be shown up as `get_attr` but a
exir/passes/constant_prop_pass.py:345
↓ 19 callersMethodduplicate_tensor
Create a new TFLite tensor, which is an identical copy of 'tensor', with a new name. If 'new_name' is given, it will be used as the name
backends/nxp/backend/ir/converter/builder/model_builder.py:308
↓ 19 callersMethodencode
Encodes a string into a list of token IDs. Args: s (str): The input string to be encoded. bos (bool): Whethe
examples/models/llama/tokenizer/tiktoken.py:119
↓ 19 callersFunctiongenerate_etrecord
Generates an `ETRecord` from the given objects, serializes it and saves it to the given path. The objects that will be serialized to an `ETRe
devtools/etrecord/_etrecord.py:561
↓ 19 callersFunctionget_lowered_submodules
Returns a list of lowered modules that are in the given graph (does not look into submodules). Specifically, the returned value is a list con
exir/lowered_backend_module.py:929
↓ 19 callersFunctionget_symmetric_a8w4_quantization_config
( is_per_channel: bool = True, is_qat: bool = True, is_dynamic: bool = False )
backends/arm/quantizer/arm_quantizer.py:289
↓ 19 callersMethodhas_compatible_packed_dim_info_set
Check if all PackedDimInfos from other are contained within self's PackedDimInfo sets, i.e. self is a superset of other for both buff
backends/vulkan/utils.py:1043
↓ 19 callersFunctiontorch_dtype_to_scalar_type
Convert torch dtype to ET ScalarType int value.
backends/mlx/builder/op_helpers.py:363
↓ 19 callersMethodupdate
( self, input_pos: torch.Tensor, k_val: torch.Tensor, v_val: torch.Tensor )
examples/models/llama/attention.py:96
↓ 19 callersMethodupdate
( self, input_pos: torch.Tensor, k_val: torch.Tensor, v_val: torch.Tensor )
examples/models/llama/source_transformation/sdpa.py:455
↓ 19 callersMethodverify_nop_memory_alloc
(self, graph_module: torch.fx.GraphModule)
backends/cadence/aot/tests/test_memory_passes.py:237
↓ 18 callersMethod__init__
(self, in_features=4, out_features=3)
backends/cortex_m/test/ops/test_activation.py:29
↓ 18 callersMethod__init__
(self)
backends/apple/mps/test/test_mps_models.py:64
↓ 18 callersFunction_annotate_nodes
(node_tuples, quant_spec, input_node=False)
backends/example/example_operators/utils.py:23
↓ 18 callersMethodaddConst
(self, shape, dtype, values, name)
backends/arm/test/misc/test_process_node.py:37
↓ 18 callersMethodcalculate_numeric_gap
Compares logged intermediate outputs from the exported graph (in ETRecord) with runtime outputs (in ETDump) using a user-specific num
devtools/inspector/_inspector.py:1570
↓ 18 callersMethodcheckShape
(long[] shape)
extension/android/executorch_android/src/main/java/org/pytorch/executorch/Tensor.java:1045
↓ 18 callersMethodcheck_op_counts
( self, graph_module: torch.fx.GraphModule, expected_op_counts: dict[EdgeOpOverload, i
backends/cadence/aot/tests/test_fusion_ops_passes.py:87
↓ 18 callersMethodcreate_unary_node
( self, node: torch.fx.Node, mps_graph: MPSGraph, unary_op: MPSNodeUnion )
backends/apple/mps/operators/node_visitor.py:372
↓ 18 callersFunctionensure_type
Ensure that the argument is of the expected type. Args: expected_type (type[T]): The expected type. arg (object): The argument to
backends/arm/common/type.py:12
↓ 18 callersMethodexecute
Executes the method with the given inputs. Args: inputs: A sequence of input values, typically torch.Tensor objects. Ret
runtime/__init__.py:138
↓ 18 callersMethodgen_tflite
Generates TFLite code for the Shape
backends/nxp/backend/ir/tflite_generator/tflite_model.py:352
↓ 18 callersFunctionget_capture_program_passes
Defines and returns the default ordered passes for the capture program. This function creates an OrderedDict containing a series of default p
backends/qualcomm/_passes/qnn_pass_manager.py:90
↓ 18 callersFunctionget_conv2d_output_size
( in_size: torch.Size, out_channels: int, stride: Tuple[int], padding: Tuple[int], dilatio
backends/cadence/aot/utils.py:86
↓ 18 callersMethodget_inputs
(self)
backends/arm/test/ops/test_ge.py:42
↓ 18 callersMethodget_inputs
(self)
backends/arm/test/ops/test_le.py:43
↓ 18 callersMethodget_inputs
(self)
backends/arm/test/ops/test_lt.py:43
↓ 18 callersMethodget_inputs
(self)
backends/arm/test/ops/test_gt.py:43
↓ 18 callersMethodget_scale_per_tensor
(self)
backends/arm/_passes/quant_args.py:101
↓ 18 callersMethodmake_tmp_value_slot
Create a temporary SymInt slot and register it.
backends/mlx/builder/slot_manager.py:140
↓ 18 callersMethodregister_backend_recipe_provider
Register a backend recipe provider
export/recipe_registry.py:37
↓ 18 callersMethodrun
(self)
setup.py:691
↓ 18 callersFunctiontap_intermediate_outputs_
Rewrite `ep` IN PLACE so each node matching a rule's selector has its output appended to the program outputs (wrapped in a `tap.Tensor` p
devtools/intermediate_output_tap/_tap_pass.py:57
↓ 18 callersMethodtransform
(self)
backends/arm/test/tester/arm_tester.py:277
↓ 18 callersFunctionvalidate
Validate that two graph modules produce numerically equivalent outputs. Args: original: The original graph module before the pass
backends/cadence/aot/tests/test_remove_ops_passes.py:47
↓ 17 callersMethodInitFromBuf
(cls, buf, pos)
exir/_serialize/generated/executorch_flatbuffer/Int.py:65
↓ 17 callersMethod_assign_format_to_node
Assign format to node, but only if it's not channels first.
backends/nxp/backend/node_format_inference.py:181
↓ 17 callersMethod_from_list
(cls, compile_specs: list[CompileSpec])
backends/arm/common/arm_compile_spec.py:78
↓ 17 callersMethod_gen_sample_profile_event
Helper for generating test ProfileEvents Notably: - the timestamp is specified as a tuple of two separate integers -
devtools/inspector/tests/event_blocks_test.py:27
↓ 17 callersMethod_run_test
( self, B, H_q, H_kv, Lq, Lk, D, attn_mask=None, is_causal=False, min_cosine=0.95 )
backends/cuda/tests/test_tq4_sdpa.py:150
↓ 17 callersMethodarguments
(self, *, include_context: bool = True)
codegen/api/types/signatures.py:36
↓ 17 callersMethodcompare_partitions
Compare two sets of partitions to see if they are equivalent. Two sets of partitions are considered equivalent if: 1. They ha
exir/backend/test/test_group_partitioner.py:458
↓ 17 callersMethodconvert
Replaces the given observer node from the given model with a quantized weight and a OpenVINO specific decompression module.
backends/openvino/quantizer/observers.py:103
↓ 17 callersFunctiondiff_pte
( data_a: bytes, data_b: bytes, path_a: str = "A", path_b: str = "B", max_samples: int = 1
devtools/pte_tool/diff_pte.py:480
↓ 17 callersFunctiondim_order_from_stride
Dimension order represents how dimensions are laid out in memory, starting from the outer-most to the inner-most dimension. Thus, the con
exir/tensor.py:54
↓ 17 callersMethoddtype
@return data type of this tensor.
extension/android/executorch_android/src/main/java/org/pytorch/executorch/Tensor.java:501
↓ 17 callersMethodembedding
(self)
examples/models/voxtral_tts/model.py:1232
↓ 17 callersMethodencode
(self, text, bos, eos)
examples/apple/coreml/llama/run.py:35
↓ 17 callersMethodfind_nodes
Return nodes of the graph module depending on NodeFinder type. Args: model (GraphModule): The graph module to search for matching
backends/arm/quantizer/arm_quantizer_utils.py:226
↓ 17 callersFunctiongenerate_pattern_op_partitions
Args: graph_module: Module that we want to partition patterns: A list of patterns in the form of torch.fx.Graph. These graphs
exir/backend/canonical_partitioners/pattern_op_partitioner.py:107
↓ 17 callersMethodget_example_inputs
( self, num_token: int = 128, cache_size: int = 512, get_dym_shape=False )
examples/mediatek/models/llm_models/modeling_common.py:818
↓ 17 callersFunctionget_param_tensor
( exp_prog: ExportedProgram, node: torch.fx.Node )
backends/xnnpack/utils/utils.py:108
↓ 17 callersFunctionget_soc_to_chipset_map
()
backends/qualcomm/utils/utils.py:1305
↓ 17 callersMethodget_zp_per_tensor
(self)
backends/arm/_passes/quant_args.py:109
↓ 17 callersFunctionquantize_model
Walk model parameters + persistent buffers, apply recipe. Returns a single state dict containing quantized tensor subclasses (``Int4Tensor``,
examples/models/gemma4_31b/quant/quantize.py:282
↓ 17 callersMethodrun
( self, artifact: ExportedProgram, inputs=None, generate_etrecord: bool = False )
backends/qualcomm/tests/tester.py:75
↓ 17 callersFunctionrun_test
Top-level test run function for a model, input set, and tester. Handles test execution and reporting.
backends/test/suite/runner.py:87
↓ 17 callersMethodrun_upsample_aa_test
Test our ExecuTorch anti-aliased bilinear upsampling against PyTorch reference.
kernels/portable/test/op_upsample_bilinear2d_aa_test.py:36
↓ 17 callersFunctionselect_by_op_type
Match nodes whose `str(node.target)` ends with any of `op_targets`. The "ends with" rule lets the user write either the short name ("ate
devtools/intermediate_output_tap/_selectors.py:51
↓ 17 callersMethodto_out_variant
Find out the out-variant of this operator and return it. TODO (larryliu): Implement execution dialect class and let this function return that.
exir/dialects/edge/_ops.py:305
↓ 17 callersFunctionvalidate_numerics
Validate that two graph modules produce numerically equivalent outputs. Args: original: The original graph module before the pass
backends/cadence/aot/tests/test_fusion_ops_passes.py:47
↓ 16 callersMethod__init__
(self)
test/end2end/test_end2end.py:107
↓ 16 callersMethod__init__
( self, in_channels=2, out_channels=1, kernel_size=(3, 3), stride=(2,
backends/xnnpack/test/ops/test_conv2d.py:38
↓ 16 callersMethod__init__
(self)
backends/apple/mps/test/test_mps_binary_ops.py:14
↓ 16 callersMethod_apply_pass_insertions
Apply all registered pass insertions to the collected passes. Called ONCE after all add_passes() calls are complete, before execution.
backends/arm/_passes/arm_pass_manager.py:305
↓ 16 callersMethod_is_quantized_meta
Return True when meta indicates fully quantized inputs and outputs.
backends/arm/_passes/arm_pass.py:43
↓ 16 callersFunction_max_abs_error
(out, ref)
backends/cuda/tests/test_triton_sdpa.py:66
↓ 16 callersFunction_reference_sdpa
Compute reference SDPA in float32 with expanded KV heads for GQA. Adapted from FlashAttention's testing.py: expand KV heads via repeat_interl
backends/cuda/tests/test_triton_sdpa.py:36
↓ 16 callersMethodcheck_node_count
(self, input: Dict[Any, int])
backends/test/harness/tester.py:289
← previousnext →301–400 of 28,493, ranked by callers