mirror of
https://github.com/tinygrad/tinygrad.git
synced 2026-04-07 03:00:26 -04:00
7170a9a057
* wip mistral
* coder
* touchups
* cleanups
* mistral cleanups
* clean up cache create
* download the weights, fix tests
* fix llama loading
* global fixup
* clean up all
* move llama model
* cleanups
* Revert "cleanups"
This reverts commit a71c5d59eb.
* fine, leave it
45 lines
1.5 KiB
Python
45 lines
1.5 KiB
Python
#!/usr/bin/env python
|
|
import unittest
|
|
import numpy as np
|
|
from tinygrad.tensor import Tensor
|
|
from tinygrad.jit import TinyJit
|
|
from tinygrad.helpers import dtypes, CI
|
|
from tinygrad.ops import Device
|
|
from test.helpers import derandomize_model
|
|
|
|
from examples.llama import Transformer
|
|
|
|
def helper_test_jitted_correctness(gen, train, train_jit):
|
|
nojit = train(*gen()).numpy()
|
|
for _ in range(5): jit = train_jit(*gen()).numpy()
|
|
np.testing.assert_allclose(nojit, jit, rtol=1e-3, atol=1e-5)
|
|
|
|
class TestJittedModels(unittest.TestCase):
|
|
def test_jitted_tiny_llama(self):
|
|
old_type = Tensor.default_type
|
|
Tensor.default_type = dtypes.float16
|
|
|
|
args_tiny = {"dim": 1024, "hidden_dim": 1024, "n_heads": 8, "n_layers": 8, "norm_eps": 1e-05, "vocab_size": 1000}
|
|
model = Transformer(**args_tiny)
|
|
derandomize_model(model)
|
|
def test(t): return model(t, 0).realize()
|
|
|
|
@TinyJit
|
|
def test_jit(t): return model(t, 0).realize()
|
|
helper_test_jitted_correctness(lambda: (Tensor([[1,]]),), test, test_jit)
|
|
Tensor.default_type = old_type
|
|
|
|
@unittest.skipUnless(not CI, "huge for CI")
|
|
def test_jitted_stable_diffusion(self):
|
|
from examples.stable_diffusion import UNetModel
|
|
model = UNetModel()
|
|
derandomize_model(model)
|
|
def test(t, t2): return model(t, 801, t2).realize()
|
|
|
|
@TinyJit
|
|
def test_jit(t, t2): return model(t, 801, t2).realize()
|
|
helper_test_jitted_correctness(lambda: (Tensor.randn(1, 4, 16, 16),Tensor.randn(1, 77, 768)), test, test_jit)
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|