Files
github-actions[bot] eefc084302 sunnypilot v2025.12.18-3820
version: sunnypilot v2025.003.000 (dev)
date: 2025-12-18T05:48:43
master commit: 16c052af0835f049a67b80251d8cc1114fc08bb4
2025-12-18 05:48:43 +00:00

18 lines
414 B
Python

from tinygrad import Tensor, TinyJit, Device
import numpy as np
GPUS = 4
N = 128
ds = tuple([Device.canonicalize(f"{Device.DEFAULT}:{i}") for i in range(GPUS)])
t = Tensor.rand(N, N, N).shard(ds, 0)
n = t.numpy()
@TinyJit
def allreduce(t:Tensor) -> Tensor:
return t.sum(0) #.realize()
for i in range(10):
print(i)
tn = allreduce(t).numpy()
np.testing.assert_allclose(tn, n.sum(0), atol=1e-4, rtol=1e-4)