mirror of
https://github.com/tinygrad/tinygrad.git
synced 2026-06-13 08:28:55 +08:00
a9a1df785f
* initial commit * 81 passing * 105 passing tests * 148 passing * CI tests * install dep on ci * try opencl pkgs * try using vulkan * down to only 6 failing * refactor * cleaning up * another test skipped due to buffer limit * linter * segfault * indent fix * another segfault found * small touchups * Fix max and maxpool tests * Add constant folding * Add javascript export script * better asserts in codegen * manual upcasting * reverted token type change * skip safetensor test due to unsupported type * FIx efficientnet and all other model tests * Remove np copy * fixed indent and missing import * manually destroy the buffer * revert back to length * linter errors * removed extra val * skip broken tests * skipping more tests * Make the page pretty * Save model weights as safetensor * Fix imagenet to c test * Fix second imagenet to c bug * Async and paralel kernel compilation * workgroup support * reversed local size * fixed non local bug * correct local groups * ci experiment * removed typo * Fix define local by using shared memory * Refactor * try running on mac * match metal tests * add more workers * scope down tests * trying windows runner * fixed windows env * see how many it can do * merged master * refactor * missed refactor * increase test suite coverage * missing import * whitespace in test_efficientnet.py * getting there * fixed reset * fixed bufs * switched to cstyle * cleanup * min/max rename * one more linter issue * fixed demo * linter * testing ci chrome * add unsafe webgpu arg * add build step * remove WEBGPU from cmd line * use module * try forcing directx * trying forced metal backend * temp disable conv2d for CI * disable conv_trasnpose2d --------- Co-authored-by: 0x4d - Martin Loretz <20306567+martinloretzzz@users.noreply.github.com> Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
54 lines
2.2 KiB
Python
54 lines
2.2 KiB
Python
import unittest
|
|
from tinygrad.tensor import Tensor
|
|
from tinygrad.helpers import dtypes
|
|
from tinygrad.lazy import Device
|
|
# similar to test/external/external_test_gpu_ast.py, but universal
|
|
|
|
class TestSpecific(unittest.TestCase):
|
|
# from openpilot
|
|
|
|
# 1x1 6 <- 24
|
|
def test_1x1_6_24(self):
|
|
x = Tensor.randn(1, 24*4, 32, 64)
|
|
w = Tensor.randn(6*4, 24*4, 1, 1)
|
|
x.conv2d(w).permute(0,2,3,1).reshape(32, 384, 4).contiguous().realize()
|
|
|
|
def test_vec_mul(self):
|
|
# this forces it to be an image...
|
|
x = Tensor.ones(1, 512, 4).contiguous().reshape(1, 2048)
|
|
w = Tensor.randn(2048, 512)
|
|
(x @ w).reshape(1, 128, 4).contiguous().realize()
|
|
|
|
@unittest.skipIf(Device.DEFAULT in ["LLVM", "WEBGPU"], "Broken on LLVM and webgpu")
|
|
def test_big_vec_mul(self):
|
|
# from LLaMA
|
|
# 0 buffer<4096, dtypes.float> [View((1024, 1, 1, 4), (4, 0, 0, 1), 0, None)]
|
|
# 1 buffer<4096, dtypes.float> [View((1024, 1024, 4, 4), (0, 4, 1, 0), 0, None)]
|
|
# 2 buffer<16777216, dtypes.half> [View((1024, 1024, 4, 4), (16384, 4, 1, 4096), 0, None)]
|
|
x = Tensor.randn(4096).realize()
|
|
w = Tensor.randn(4096, 4096, device='cpu').cast(dtypes.float16).to(Device.DEFAULT).realize()
|
|
(x @ w.T).realize()
|
|
|
|
# from https://dl.acm.org/doi/pdf/10.1145/3495243.3517020
|
|
|
|
# ~260 GFLOPS on Adreno 640, should be 260*(720/890)*(596/710) = 176.5 on downclocked 630
|
|
# we get 170
|
|
def test_1x1_28_28(self):
|
|
x = Tensor.randn(1, 256, 28, 28)
|
|
w = Tensor.randn(256, 256, 1, 1)
|
|
x.conv2d(w).permute(0,2,3,1).reshape(28, 28*256//4, 4).contiguous().realize()
|
|
|
|
# 132 GFLOPS on Adreno 640, should be 132*(720/890)*(596/710) = 90 on downclocked 630
|
|
# gets 54 with broken opt, 74 without opt, and 146 if we pad and opt 3!
|
|
def test_3x3_28_28_stride_2(self):
|
|
x = Tensor.randn(1, 288, 36, 36)
|
|
w = Tensor.randn(384, 288, 3, 3)
|
|
x.conv2d(w, stride=2).permute(0,2,3,1).reshape(17, 17*384//4, 4).contiguous().realize()
|
|
|
|
def test_3x3_28_28_stride_2_padded(self):
|
|
x = Tensor.randn(1, 288, 36, 36)
|
|
w = Tensor.randn(384, 288, 3, 3)
|
|
x.conv2d(w, stride=2, padding=1).permute(0,2,3,1).reshape(18, 18*384//4, 4).contiguous().realize()
|
|
|
|
if __name__ == '__main__':
|
|
unittest.main() |