Closed bitfort closed 1 year ago
Basic question but important.
https://github.com/mlcommons/inference_policies/blob/master/inference_rules.adoc#411-constraints-for-the-closed-division Resnet50-v1.5 "99% of FP32 (76.46%)"
How is the quality target to be read? (A) [0.99] [FP32 (76.46%)] = 75.695% (B) [0.99 FP32] = 76.46%
Based on my understanding going through previous log files, (A) appears to be the correct reading but want to confirm here.
A is correct (75.695%)
Thanks!
Basic question but important.
https://github.com/mlcommons/inference_policies/blob/master/inference_rules.adoc#411-constraints-for-the-closed-division Resnet50-v1.5 "99% of FP32 (76.46%)"
How is the quality target to be read? (A) [0.99] [FP32 (76.46%)] = 75.695% (B) [0.99 FP32] = 76.46%
Based on my understanding going through previous log files, (A) appears to be the correct reading but want to confirm here.