Closed ylwu-amzn closed 2 years ago
It will be good if we can know how much memory a model will use to evaluate if we have enough memory to run a new model. AD estimates memory usage https://github.com/opensearch-project/anomaly-detection/blob/a79028f8ee2098fa94e5e4b8ba81affce24b2b8c/src/main/java/org/opensearch/ad/MemoryTracker.java#L261. Now we are going to integrate RCF in MLCommons, that means we need to copy the same code to MLCommons. How about we build memory estimation in this RCF lib directly?
No. As stated "AD estimates memory usage..." that belongs in AD. AD should perform benchmarking.
It will be good if we can know how much memory a model will use to evaluate if we have enough memory to run a new model. AD estimates memory usage https://github.com/opensearch-project/anomaly-detection/blob/a79028f8ee2098fa94e5e4b8ba81affce24b2b8c/src/main/java/org/opensearch/ad/MemoryTracker.java#L261. Now we are going to integrate RCF in MLCommons, that means we need to copy the same code to MLCommons. How about we build memory estimation in this RCF lib directly?