github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-01-09 15:08:02 -05:00
Code Issues Packages Projects Releases Wiki Activity

replace layer_init_uniform with .uniform

Browse Source
This commit is contained in:
George Hotz
2020-12-06 13:44:31 -08:00
parent c71a8ef222
commit 102e6356e9
5 changed files with 15 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
README.md
View File

@@ -57,12 +57,11 @@ It turns out, a decent autograd tensor library is 90% of what you need for neura
```python
from tinygrad.tensor import Tensor
import tinygrad.optim as optim
from tinygrad.utils import layer_init_uniform
class TinyBobNet:
def __init__(self):
self.l1 = Tensor(layer_init_uniform(784, 128))
self.l2 = Tensor(layer_init_uniform(128, 10))
self.l1 = Tensor.uniform(784, 128)
self.l2 = Tensor.uniform(128, 10)
def forward(self, x):
return x.dot(self.l1).relu().dot(self.l2).logsoftmax()

7
examples/train_efficientnet.py
View File

@@ -4,7 +4,6 @@ import numpy as np
from extra.efficientnet import EfficientNet
from tinygrad.tensor import Tensor
from tinygrad.utils import get_parameters, fetch
from tinygrad.utils import layer_init_uniform
from tqdm import trange
import tinygrad.optim as optim
import io
@@ -15,9 +14,9 @@ class TinyConvNet:
def __init__(self, classes=10):
conv = 3
inter_chan, out_chan = 8, 16 # for speed
self.c1 = Tensor(layer_init_uniform(inter_chan,3,conv,conv))
self.c2 = Tensor(layer_init_uniform(out_chan,inter_chan,conv,conv))
self.l1 = Tensor(layer_init_uniform(out_chan*6*6, classes))
self.c1 = Tensor.uniform(inter_chan,3,conv,conv)
self.c2 = Tensor.uniform(out_chan,inter_chan,conv,conv)
self.l1 = Tensor.uniform(out_chan*6*6, classes)
def forward(self, x):
x = x.conv2d(self.c1).relu().max_pool2d()

12
test/test_mnist.py
View File

@@ -3,7 +3,7 @@ import os
import unittest
import numpy as np
from tinygrad.tensor import Tensor, GPU
from tinygrad.utils import layer_init_uniform, fetch
from tinygrad.utils import fetch
import tinygrad.optim as optim
from tqdm import trange
@@ -23,8 +23,8 @@ X_train, Y_train, X_test, Y_test = fetch_mnist()
# create a model
class TinyBobNet:
def __init__(self):
self.l1 = Tensor(layer_init_uniform(784, 128))
self.l2 = Tensor(layer_init_uniform(128, 10))
self.l1 = Tensor.uniform(784, 128)
self.l2 = Tensor.uniform(128, 10)
def parameters(self):
return [self.l1, self.l2]
@@ -39,9 +39,9 @@ class TinyConvNet:
conv = 3
#inter_chan, out_chan = 32, 64
inter_chan, out_chan = 8, 16 # for speed
self.c1 = Tensor(layer_init_uniform(inter_chan,1,conv,conv))
self.c2 = Tensor(layer_init_uniform(out_chan,inter_chan,conv,conv))
self.l1 = Tensor(layer_init_uniform(out_chan*5*5, 10))
self.c1 = Tensor.uniform(inter_chan,1,conv,conv)
self.c2 = Tensor.uniform(out_chan,inter_chan,conv,conv)
self.l1 = Tensor.uniform(out_chan*5*5, 10)
def parameters(self):
return [self.l1, self.c1, self.c2]

4
tinygrad/tensor.py
View File

@@ -117,6 +117,10 @@ class Tensor:
def randn(*shape, **kwargs):
return Tensor(np.random.randn(*shape).astype(np.float32), **kwargs)
@staticmethod
def uniform(*shape, **kwargs):
return Tensor((np.random.uniform(-1., 1., size=shape)/np.sqrt(np.prod(shape))).astype(np.float32), **kwargs)
@staticmethod
def eye(dim, **kwargs):
return Tensor(np.eye(dim).astype(np.float32), **kwargs)

4
tinygrad/utils.py
View File

@@ -1,10 +1,6 @@
import numpy as np
from tinygrad.tensor import Tensor
def layer_init_uniform(*x):
ret = np.random.uniform(-1., 1., size=x)/np.sqrt(np.prod(x))
return ret.astype(np.float32)
def fetch(url):
import requests, os, hashlib, tempfile
fp = os.path.join(tempfile.gettempdir(), hashlib.md5(url.encode('utf-8')).hexdigest())