buffer comparator difference

[jimmylihui]

Have anyone met this problem before?

Benchmarking sequential method: 0.00010 seconds
2023-12-09 15:50:44.287478: E external/xla/xla/service/gpu/buffer_comparator.cc:143] Difference at 16: 1.85469, expected 0.588866
2023-12-09 15:50:44.287548: E external/xla/xla/service/gpu/buffer_comparator.cc:143] Difference at 17: 2.10676, expected 1.64956
2023-12-09 15:50:44.287568: E external/xla/xla/service/gpu/buffer_comparator.cc:143] Difference at 19: 1.34889, expected 1.0033
2023-12-09 15:50:44.287584: E external/xla/xla/service/gpu/buffer_comparator.cc:143] Difference at 20: 1.54572, expected 0.773441
2023-12-09 15:50:44.287599: E external/xla/xla/service/gpu/buffer_comparator.cc:143] Difference at 21: 1.84754, expected 2.27585
2023-12-09 15:50:44.287614: E external/xla/xla/service/gpu/buffer_comparator.cc:143] Difference at 22: 1.01156, expected 1.69631
2023-12-09 15:50:44.287629: E external/xla/xla/service/gpu/buffer_comparator.cc:143] Difference at 23: 0.835973, expected 1.37905
2023-12-09 15:50:44.287644: E external/xla/xla/service/gpu/buffer_comparator.cc:143] Difference at 24: 1.37483, expected 0.50073
2023-12-09 15:50:44.287659: E external/xla/xla/service/gpu/buffer_comparator.cc:143] Difference at 26: 1.03055, expected 0.796119
2023-12-09 15:50:44.287674: E external/xla/xla/service/gpu/buffer_comparator.cc:143] Difference at 28: 0.406476, expected 0.849054
E1209 15:50:44.287688   42926 triton_autotuner.cc:766] Results do not match the reference. This is likely a bug/unexpected loss of precision.

The code is in https://github.com/jimmylihui/deer. please run the demo_2.py in deer folder

[JankowskiChristopher]

Yes, I've just had this problem.

[btlizard]

I had this problem too, when I set my environment larger.

[yanboyang97]

have you solved it yet?

[JankowskiChristopher]

I unfortunately did not solve it and had to switch to PyTorch instead of using JAX for the project I had been involved in.