Created
August 27, 2025 18:25
-
-
Save rebeccajae/f1767b9ced719aee03a073977c60245b to your computer and use it in GitHub Desktop.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| import torch | |
| def main(): | |
| if not torch.backends.mps.is_available(): | |
| print("MPS not available") | |
| return | |
| large_weight = torch.randn(12, 8, device='mps') | |
| weight_sliced = large_weight[::2, ::1] | |
| weight_contiguous_equiv = weight_sliced.contiguous() | |
| input_s = torch.randn(2, 8, device='mps') | |
| result_sliced = torch.nn.functional.linear(input_s, weight_sliced) | |
| result_contig = torch.nn.functional.linear(input_s, weight_contiguous_equiv) | |
| input_cpu = input_s.cpu() | |
| weight_sliced_cpu = weight_sliced.cpu() | |
| weight_contig_cpu = weight_contiguous_equiv.cpu() | |
| result_sliced_cpu = torch.nn.functional.linear(input_cpu, weight_sliced_cpu) | |
| result_contig_cpu = torch.nn.functional.linear(input_cpu, weight_contig_cpu) | |
| print(f"weight_sliced.shape: {weight_sliced.shape}") | |
| print(f"input_s.shape: {input_s.shape}") | |
| print(f"weight_sliced.is_contiguous(): {weight_sliced.is_contiguous()}") | |
| print(f"weight_sliced.stride(): {weight_sliced.stride()}") | |
| print(f"weight_contiguous_equiv.stride(): {weight_contiguous_equiv.stride()}") | |
| print(f"torch.equal(weight_sliced, weight_contiguous_equiv): {torch.equal(weight_sliced, weight_contiguous_equiv)}") | |
| print(f"MPS: Contiguous and non-contiguous match: {torch.allclose(result_contig, result_sliced, atol=1e-4)}") | |
| print(f"CPU: Contiguous and non-contiguous match: {torch.allclose(result_contig_cpu, result_sliced_cpu, atol=1e-4)}") | |
| print(f"MPS contiguous vs CPU contiguous: {torch.allclose(result_contig, result_contig_cpu.to('mps'), atol=1e-4)}") | |
| print(f"MPS non-contig vs CPU non-contig: {torch.allclose(result_sliced, result_sliced_cpu.to('mps'), atol=1e-4)}") | |
| mps_matches_internally = torch.allclose(result_contig, result_sliced, atol=1e-4) | |
| mps_matches_cpu = torch.allclose(result_sliced, result_sliced_cpu.to('mps'), atol=1e-4) | |
| if not mps_matches_internally: | |
| print("MPS result not consistent") | |
| if not mps_matches_cpu: | |
| print("MPS result does not match CPU result") | |
| print(f"MPS contiguous result sample: {result_contig.flatten()[:5]}") | |
| print(f"MPS non-contig result sample: {result_sliced.flatten()[:5]}") | |
| print(f"CPU non-contig result sample: {result_sliced_cpu.flatten()[:5]}") | |
| if __name__ == "__main__": | |
| main() |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment