Pablo Gonzalez

Results 45 comments of Pablo Gonzalez

#301 contains the updated accuracy numbers

This error seems very weird, since if you set `performance_sample_count_override`, the `performance_sample_count` should be the size of the dataset. https://github.com/mlcommons/inference/blob/c4a19872d9e7ba2fe2d5b8a4c3d3c02e82233785/loadgen/test_settings_internal.cc#L117-L119 For 3d-unet this should be 43. You could try to...

@ShriyaPalsamudram I added the flag to skip the seed checker. @hiwotadese for visibility

@arjunsuresh How should we handle when a system is renamed? I see a couple of options here, but not sure if they are a what we are looking for -...

@ColaDrill What were the RAM and GPUs of the machine you tried to run the benchmark on?

@arjunsuresh I was not familiar with this issue. But I am aware of the urgency of this issue and I am currently investigating it: - How much ram does your...

@arjunsuresh Here is the logs of an accuracy run I made when testing the reference implementation: [link](https://drive.google.com/drive/folders/10woMZaptlE2Q89DNQHEgkuYaDwIZ34jc?usp=drive_link ) The only file missing `mlperf_log_accuracy.json`, that is about 1.3GB. If you think...

Solution for 4.0: https://github.com/mlcommons/logging/tree/training_v4.0