Pinzhen Xu pinzhenx

OS @apple | Former ML @alibaba @intel | PyTorch

pinzhenx / memstat.py

Created February 19, 2022 08:09

How is MemAvailable calculated in Linux? Python version of the kernel calculating the memory stats

	pgsize = 4 # kb
	total_lowwmark = 0
	total_reserved = 0
	with open('/proc/zoneinfo') as zoneinfo:
	for line in zoneinfo:
	if 'pages free' in line:
	next(zoneinfo)
	low = int(next(zoneinfo).split()[1])
	high = int(next(zoneinfo).split()[1])
	total_lowwmark += low

pinzhenx / conv_denorm.ipynb

Last active November 5, 2021 14:01

conv_denorm.ipynb

Sorry, something went wrong. Reload?

Sorry, we cannot display this file.

Sorry, this file is invalid so it cannot be displayed.

pinzhenx / mirror.ipynb

Last active August 16, 2021 07:17

Anatomy of mirror strategy

Sorry, something went wrong. Reload?

Sorry, we cannot display this file.

Sorry, this file is invalid so it cannot be displayed.

pinzhenx / benchmark_conv.py

Last active December 7, 2020 06:41

	import torch
	import pandas as pd

	def profile(m, x, nwarm=10, nrun=300):
	for _ in range(nwarm):
	m(x)
	with torch.autograd.profiler.profile(True) as prof:
	for _ in range(nrun):
	m(x)
	return getattr(prof.key_averages()[0], 'cpu_time') / 1000

pinzhenx / pyrys.md

Last active April 26, 2020 04:45

    A simple example of how const prop works

def foo(x):
    a = 1 + 2
    b = a + 3
    c = b + 4
    return x + c

jit_foo = torch.jit.script(foo)

pinzhenx / test_onnx.html

Created May 3, 2019 17:06

onnx sample

	<html>
	<script src='../dist/webml-polyfill.js'></script>
	<script src='third_party/protobuf.min.js'></script>
	<script src='util/base.js'></script>
	<script src='util/onnx/onnx.js'></script>
	<script src='util/onnx/OnnxModelUtils.js'></script>
	<script src='util/onnx/OnnxModelImporter.js'></script>
	<script>
	(async () => {
	const res = await fetch('path/to/model.onnx');