github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-04-29 03:00:14 -04:00
Code Issues Packages Projects Releases Wiki Activity

realize sched

Browse Source
This commit is contained in:
George Hotz
2025-10-14 14:44:56 +08:00
parent fe683bafa6
commit 30ff87eab4
1 changed files with 2 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
test/test_outerworld.py
View File

@@ -19,10 +19,9 @@ class TestOuterworldTrain(unittest.TestCase):
steps = UOp.range(16, -1)
opt.zero_grad()
loss = (layer(X[steps]) - Y[steps]).square().mean().backward()
opt.schedule_step() # TODO: does this need to know anything about steps?
sched = opt.schedule_step() # TODO: does this need to know anything about steps?
# NOTE: this can't work. the inputs to layer are not the assign, need to run twice for the fixed point?
all_losses = loss.reshape(1).expand(steps).contiguous()
all_losses.realize()
all_losses = Tensor.realize(loss.reshape(1).expand(steps).contiguous(), *sched)
print(all_losses.numpy())
#@unittest.skip("TODO: understand assign")