dhruvbird’s gists

dhruvbird / conv2d_linear_equivalence.py

Created June 20, 2023 14:44

Showing equivalence of nn.Conv2d and nn.Linear

	import torch
	from torch import nn
	from torch.nn import functional as F
	from matplotlib import pyplot as plt

	n_out_channels = 1
	mat = torch.arange(0, 36).reshape((6, 6)).float()
	print(f"mat:\n{mat}")

	patches = []

dhruvbird / selected_operators.yaml

Created September 9, 2022 23:52

	include_all_non_op_selectives: false
	build_features: []
	operators:
	aten::__getitem__.t:
	is_used_for_training: false
	is_root_operator: true
	include_all_overloads: false
	aten::_set_item.str:
	is_used_for_training: false
	is_root_operator: true

dhruvbird / FetchModelMetadata.cpp

Created November 16, 2021 23:37

Fetch metadata only using PyTorch Lite Model C++ API

	#include <torch/csrc/jit/mobile/import.h>
	#include <torch/csrc/jit/mobile/module.h>

	int main() {
	std::string model_path = "AddTensorsModel.ptl";
	torch::jit::ExtraFilesMap extra_files;

	// Add keys into extra_files to indicate which metadata files
	// we wish to fetch.
	extra_files["model_info.txt"] = "";

dhruvbird / ListArchiveContents.py

Created November 16, 2021 22:35

Print the contents of the model file and the contents of the newly added metadata file

	zf = zipfile.ZipFile("AddTensorsModel.ptl")
	zf.infolist()

	[<ZipInfo filename='AddTensorsModel/extra/model_info.txt' file_size=70>,
	<ZipInfo filename='AddTensorsModel/data.pkl' file_size=86>,
	<ZipInfo filename='AddTensorsModel/code/__torch__.py' compress_type=deflate file_size=247 compress_size=166>,
	<ZipInfo filename='AddTensorsModel/code/__torch__.py.debug_pkl' file_size=145>,
	<ZipInfo filename='AddTensorsModel/constants.pkl' file_size=4>,
	<ZipInfo filename='AddTensorsModel/bytecode.pkl' file_size=452>,
	<ZipInfo filename='AddTensorsModel/version' file_size=2>,

dhruvbird / AddMetadataToPyTorchLiteModel.py

Created November 16, 2021 22:33

Add additional metadata to PyTorch Lite Model

	zf2 = zipfile.ZipFile("AddTensorsModel.ptl", "a")
	contents = "Both tensors can be any dtype as long as they can be added"
	zf2.writestr("{}/extra/dtype_info.txt".format(archive_name), contents)
	zf2.close()

dhruvbird / GetArchiveName.py

Created November 16, 2021 22:28

Get the top level folder name of the PyTorch Lite Model

	elem = zf.infolist()[0]
	archive_name = elem.filename.split("/")[0]
	print(archive_name)

	AddTensorsModel

dhruvbird / ExtractFileFromZip.py

Created November 16, 2021 17:37

Use the zipfile module to view the contents of the added metadata file

	print(zf.read("AddTensorsModel/extra/model_info.txt"))

	b"This model's forward() method accepts 2 tensors, and returns their sum"

dhruvbird / ViewModelAsZipFile.py

Created November 16, 2021 17:34

View the contents of a PyTorch Lite Model

	zf = zipfile.ZipFile("AddTensorsModel.ptl")
	zf.infolist()


	[<ZipInfo filename='AddTensorsModel/extra/model_info.txt' file_size=70>,
	<ZipInfo filename='AddTensorsModel/data.pkl' file_size=86>,
	<ZipInfo filename='AddTensorsModel/code/__torch__.py' compress_type=deflate file_size=247 compress_size=166>,
	<ZipInfo filename='AddTensorsModel/code/__torch__.py.debug_pkl' file_size=145>,
	<ZipInfo filename='AddTensorsModel/constants.pkl' file_size=4>,
	<ZipInfo filename='AddTensorsModel/bytecode.pkl' file_size=452>,

dhruvbird / CreateSimplePyTorchModelAndSaveInLiteFormat.py

Created November 16, 2021 17:33

Create a simple PyTorch Model and save it in Lite Format along with a single metadata file (model_info.txt)

	import torch
	import zipfile

	class AddTensorsModel(torch.nn.Module):
	def __init__(self):
	super().__init__();

	def forward(self, x, y):
	return x + y

dhruvbird / ConstantSpaceInOrderTraversalBSTArrayRepresentation.py

Created November 14, 2021 03:02

Code showing the algorithm for Constant space, Linear Time In-Order Traversal of a Binary Tree stored in an Array in Heap Order

	def fill_inorder_recursive_bst_array(bst, idx, val):
	"""Recursively fill an array "bst" with values so that values
	increment by 1 every time. "idx" is the index in "bst" that
	the method should look up to write out the next value in sequence.

	The first invocation is expected to pass idx == 0, with enough
	space in the BST array (bst) to hold a complete and balanced
	binary tree.

	This method returns the next value in sequence that need to be

Dhruv Matani dhruvbird