|
| 1 | +# Adapted from https://github.com/thinking-machines-lab/batch_invariant_ops/blob/main/batch_invariant_ops/test_batch_invariance.py |
| 2 | + |
| 3 | +import unittest |
| 4 | + |
| 5 | +import paddle |
| 6 | + |
| 7 | +from custom_ops.batch_invariant_ops import set_batch_invariant_mode |
| 8 | + |
| 9 | + |
| 10 | +class TestBatchInvariantForMM(unittest.TestCase): |
| 11 | + def setUp(self): |
| 12 | + """ |
| 13 | + Initialize the test environment |
| 14 | + """ |
| 15 | + device = "gpu" if paddle.is_compiled_with_cuda() else "cpu" |
| 16 | + paddle.set_device(device) |
| 17 | + |
| 18 | + def test_batch_invariance(self, B: int = 2048, D: int = 4096, dtype=paddle.float32): |
| 19 | + a = paddle.linspace(-100, 100, B * D, dtype=dtype).reshape(B, D) |
| 20 | + b = paddle.linspace(-100, 100, D * D, dtype=dtype).reshape(D, D) |
| 21 | + |
| 22 | + # Method 1: Matrix-vector multiplication (batch size 1) |
| 23 | + out1 = paddle.mm(a[:1], b) |
| 24 | + |
| 25 | + # Method 2: Matrix-matrix multiplication, then slice (full batch) |
| 26 | + out2 = paddle.mm(a, b)[:1] |
| 27 | + |
| 28 | + # Check if results are identical |
| 29 | + diff = (out1 - out2).abs().max() |
| 30 | + return diff.item() == 0, diff |
| 31 | + |
| 32 | + def run_iters(self, iters=10, ass=False): |
| 33 | + for dtype in [paddle.float32, paddle.bfloat16]: |
| 34 | + is_deterministic = True |
| 35 | + difflist = [] |
| 36 | + for i in range(iters): |
| 37 | + isd, df = self.test_batch_invariance(dtype=dtype) |
| 38 | + is_deterministic = is_deterministic and isd |
| 39 | + difflist.append(df) |
| 40 | + print( |
| 41 | + f"Batch Deterministic: {is_deterministic} run-to-run max/min/diff {max(difflist)}/{min(difflist)}/{max(difflist)-min(difflist)} for {dtype} in {iters} iterations" |
| 42 | + ) |
| 43 | + if ass: |
| 44 | + assert max(difflist) == 0 |
| 45 | + |
| 46 | + def test_case(self): |
| 47 | + # Test with standard Paddle (likely to show differences) |
| 48 | + print("Standard Paddle:") |
| 49 | + with set_batch_invariant_mode(False): |
| 50 | + self.run_iters(ass=False) |
| 51 | + # Test with batch-invariant operations |
| 52 | + print("\nBatch-Invariant Mode:") |
| 53 | + with set_batch_invariant_mode(True): |
| 54 | + self.run_iters(ass=True) |
| 55 | + |
| 56 | + |
| 57 | +if __name__ == "__main__": |
| 58 | + unittest.main() |
0 commit comments