mirror of
https://github.com/tinygrad/tinygrad.git
synced 2026-04-29 03:00:14 -04:00
5d212864b5
* Add ResNet inference test and cannon * Test with ResNet50 * test_car works with resnet fix * Add KiTS19 dataset * KiTS19: Implement iterate * No batch load for this dataset * Save results on iterate * Implement dice score * Add data prep and eval functions * Resolve shape issue * Conversion works but wrong values * Segfaults when load_from_pretrained is called * Fix segfault and assign properly * Final result generated, though very slow * Store and load final result to save time * Fix typo in finalize * Score computes * More bug fixes, dice score is very low * Working broken code * Assign output values to result * Getting a much higher score now * Fix dataset preprocessing * Mean DICE score of 88.5 * Ugh, typo * Attempt to reimplement model * Rename layers * Tiny model works, kinda * Accuracy? gone * Implement InstanceNorm and match torch * Test instance norm 2d and 3d * Combined input block with downsample block * Tiny model works, support strided convtranspose * Commands to download dataset * Clean up a bit * unet3d_v2 -> unet3d * Remove duplicated code * Oops, put tests back
64 lines
1.7 KiB
Python
64 lines
1.7 KiB
Python
# load each model here, quick benchmark
|
|
from tinygrad.tensor import Tensor
|
|
from tinygrad.helpers import GlobalCounters, getenv
|
|
import numpy as np
|
|
|
|
def test_model(model, *inputs):
|
|
GlobalCounters.reset()
|
|
model(*inputs).numpy()
|
|
# TODO: return event future to still get the time_sum_s without DEBUG=2
|
|
print(f"{GlobalCounters.global_ops*1e-9:.2f} GOPS, {GlobalCounters.time_sum_s*1000:.2f} ms")
|
|
|
|
def spec_resnet():
|
|
# Resnet50-v1.5
|
|
from models.resnet import ResNet50
|
|
mdl = ResNet50()
|
|
img = Tensor.randn(1, 3, 224, 224)
|
|
test_model(mdl, img)
|
|
|
|
def spec_retinanet():
|
|
# Retinanet with ResNet backbone
|
|
from models.resnet import ResNet50
|
|
from models.retinanet import RetinaNet
|
|
mdl = RetinaNet(ResNet50(), num_classes=91, num_anchors=9)
|
|
img = Tensor.randn(1, 3, 224, 224)
|
|
test_model(mdl, img)
|
|
|
|
def spec_unet3d():
|
|
# 3D UNET
|
|
from models.unet3d import UNet3D
|
|
mdl = UNet3D()
|
|
mdl.load_from_pretrained()
|
|
img = Tensor.randn(1, 1, 128, 128, 128)
|
|
test_model(mdl, img)
|
|
|
|
def spec_rnnt():
|
|
from models.rnnt import RNNT
|
|
mdl = RNNT()
|
|
mdl.load_from_pretrained()
|
|
x = Tensor.randn(220, 1, 240)
|
|
y = Tensor.randn(1, 220)
|
|
test_model(mdl, x, y)
|
|
|
|
def spec_bert():
|
|
from models.bert import BertForQuestionAnswering
|
|
mdl = BertForQuestionAnswering()
|
|
mdl.load_from_pretrained()
|
|
x = Tensor.randn(1, 384)
|
|
am = Tensor.randn(1, 384)
|
|
tt = Tensor(np.random.randint(0, 2, (1, 384)).astype(np.float32))
|
|
test_model(mdl, x, am, tt)
|
|
|
|
if __name__ == "__main__":
|
|
# inference only for now
|
|
Tensor.training = False
|
|
Tensor.no_grad = True
|
|
|
|
for m in getenv("MODEL", "resnet,retinanet,unet3d,rnnt,bert").split(","):
|
|
nm = f"spec_{m}"
|
|
if nm in globals():
|
|
print(f"testing {m}")
|
|
globals()[nm]()
|
|
|
|
|