Torch-Kutti

• Simple and basic , tiny autograd engine backed by numpy.
• More of an research and educational framework which is similar to Pytorch.

What to expect

• Nothing, Hopefully be updating this constantly .
• Modules , losses will be implemented.

What can be done ?

Automatic differentiation

import Torch_Kutti as tk 


a = tk.Tensor([1, 2, 3], requires_grad=True)
b = tk.Tensor([4, 5, 6], requires_grad=True)

def f(x,y):
    return (x+y) + x*2 + (x*y*3)

### calculate complex functions
z = f(a,b)
print(z)

#compute gradients
z.backward()

#get gradients of specific tensors
print(a.grad)
print(b.grad)

License

MIT