Set `requires_grad` and get `grad` in PyTorch

#python #pytorch #gradient #grad

Buy Me a Coffee☕

*Memos:

My post explains requires_grad.
My post explains how to set and get dtype.
My post explains how to set and get device.
My post explains how to set keepdim argument.
My post explains how to set out argument.

You can set requires_grad and get grad as shown below:

*Memos:

I selected some popular requires_grad argument functions such as tensor(), arange(), rand(), rand_like(), zeros(), zeros_like(), full(), full_like() and eye().
Basically, requires_grad(Optional-Default:False-Type:bool).
Basically, requires_grad= must be used.
My post explains requires_grad and backward() with tensor().

tensor(). *My post explains tensor():

import torch

my_tensor = torch.tensor(data=7., requires_grad=True)

my_tensor, my_tensor.grad
# (tensor(7., requires_grad=True), None)

my_tensor.backward()

my_tensor, my_tensor.grad
# (tensor(7., requires_grad=True), tensor(1.))

arange(). *My post explains arange():

import torch

my_tensor = torch.arange(start=5, end=15, step=3, requires_grad=True)

my_tensor, my_tensor.grad
# (tensor([7.], requires_grad=True), None)

my_tensor.backward()

my_tensor, my_tensor.grad
# (tensor([7.], requires_grad=True), tensor([1.]))

rand(). *My post explains rand():

import torch

my_tensor = torch.rand(size=(1,), requires_grad=True)

my_tensor, my_tensor.grad
# (tensor([0.0030], requires_grad=True), None)

my_tensor.backward()

my_tensor, my_tensor.grad
# (tensor([0.0913], requires_grad=True), tensor([1.]))

rand_like(). *My post explains rand_like():

import torch

my_tensor = torch.rand_like(input=torch.tensor([7.]), 
                            requires_grad=True)
my_tensor, my_tensor.grad
# (tensor([0.4687], requires_grad=True), None)

my_tensor.backward()

my_tensor, my_tensor.grad
# (tensor([0.4687], requires_grad=True), tensor([1.]))

zeros(). *My post explains zeros():

import torch

my_tensor = torch.zeros(size=(1,), requires_grad=True)

my_tensor, my_tensor.grad
# (tensor([0.], requires_grad=True), None)

my_tensor.backward()

my_tensor, my_tensor.grad
# (tensor([0.], requires_grad=True), tensor([1.]))

zeros_like(). *My post explains zeros_like():

import torch

my_tensor = torch.zeros_like(input=torch.tensor([7.]), 
                             requires_grad=True)
my_tensor, my_tensor.grad
# (tensor([0.], requires_grad=True), None)

my_tensor.backward()

my_tensor, my_tensor.grad
# (tensor([0.], requires_grad=True), tensor([1.]))

full(). *My post explains full():

import torch

my_tensor = torch.full(size=(1,), fill_value=5., requires_grad=True)

my_tensor, my_tensor.grad
# (tensor([5.], requires_grad=True), None)

my_tensor.backward()

my_tensor, my_tensor.grad
# (tensor([5.], requires_grad=True), tensor([1.]))

full_like(). *My post explains full_like():

import torch

my_tensor = torch.full_like(input=torch.tensor([7.]),
                            fill_value=5., 
                            requires_grad=True)
my_tensor, my_tensor.grad
# (tensor([5.], requires_grad=True), None)

my_tensor.backward()

my_tensor, my_tensor.grad
# (tensor([5.], requires_grad=True), tensor([1.]))

eye(). *My post explains eye():

import torch

my_tensor = torch.eye(n=1, requires_grad=True)

my_tensor, my_tensor.grad
# (tensor([[1.]], requires_grad=True), None)

my_tensor.backward()

my_tensor, my_tensor.grad
# (tensor([[1.]], requires_grad=True), tensor([[1.]]))

DEV Community

Set `requires_grad` and get `grad` in PyTorch

Top comments (0)

Read next

Build an enterprise-level financial data analysis assistant: multi-source data RAG system practice based on LangChain

IceCream: A Sweet Alternative to Print Debugging in Python

Mistral vs GPT: A Comprehensive Comparison of Leading AI Models

7 AI Open Source Libraries To Build RAG, Agents & AI Search