logging
logging copied to clipboard
MLPerf™ logging library
Currently SD rcps are using steps which decrease with increasing batch-size. This will break rcp-pruning logic. Please use samples/epochs in RCP json
I may put this in the wrong place, please let me know where the request belongs. The idea is to enable filters on all the results table as there are...
[hpc_3.0.0/common.yaml](https://github.com/mlcommons/logging/blob/master/mlperf_logging/compliance_checker/hpc_3.0.0/common.yaml) doesn't exist yet, but I want to remind in this issue the data staging has been excluded from measuring runtime in HPC, so this part: [hpc_2.0.0/common.yaml#L77-L87](https://github.com/mlcommons/logging/blob/master/mlperf_logging/compliance_checker/hpc_2.0.0/common.yaml#L77-L87) should be updated...
The training WG asked that we output a clearer error for variables that changed their names: For example: - [Here](https://github.com/mlcommons/logging/pull/290): `opt_epsilon->opt_lamb_epsilon` - [Here](https://github.com/mlcommons/logging/issues/276): `train_samples->dataset_ train_samples` and `train_samples-> total_train_samples`
Package checker currently enables RCP checker to create `scaling.json` file in results directory to normalize scores if mean epochs is faster than the RCP (#271). This PR makes creation of...
When running the package checker to check for a valid MLPerf training submission, the package_checker leaves a file scaling.json in the results/.../maskrcnn directory. This file makes the submission directory invalid...
Fix #254
training rule described in https://github.com/mlcommons/training_policies/blob/master/training_rules.adoc#91-hyperparameters require num_image_candidates to be (1000 or 2000) or (1000 * batches per chip). Current check only looks for multiple of 1000: https://github.com/mlcommons/logging/blob/master/mlperf_logging/compliance_checker/training_2.0.0/closed_maskrcnn.yaml#L64
I was aware we had this doc, but we have not been maintaining it: https://docs.google.com/document/d/1VOwh8OS4W2Ev9GkO7o_fOruL4ODECsE3/edit#heading=h.gjdgxs