mlcommons/logging

Issues

• Why is RCP pruned based on min instead of mean values?

  #367 opened 6 months ago by nv-rborkar
  5

• Correctly implement lower convex envelope in RCP pruning logic

  #368 opened 6 months ago by nv-rborkar
  0

• [v4.0 improvement] fix Stable Diffusion RCPs to use epochs/samples instead of step

  #351 opened 6 months ago by nv-rborkar
  1

• [training_4.0][gpt3][compliance checker] wrong `compile_time_mins` for gpt3 task

  #366 opened 7 months ago by ZhiyuLi-goog
  1

• [v4.0 improvement] Power result summarizer for Training/HPC should also show perf.

  #348 opened 9 months ago by nv-rborkar
  1

• [v4.0 improvement] rcp-bypass should be a per benchmark option instead of being applied to the whole submission

  #350 opened 10 months ago by nv-rborkar
  0

• [v4.0 improvement] Add node count to output of results-summarizer

  #349 opened 10 months ago by nv-rborkar
  0

• [v4.0 improvement] Restrict "status" in system.json

  #347 opened 10 months ago by nv-rborkar
  0

• Data is very sparse on the website, can filter added

  #343 opened a year ago by hanyunfan
  0

• Release of Training v3.1 package scripts

  #332 opened a year ago by keithachorn-intel
  1

• Inconsistent definition of `train_samples`

  #276 opened a year ago by hanlint
  1

• Exclude data staging from runtime in hpc_3.0.0/common.yaml checks

  #318 opened a year ago by ar-nowaczynski
  0

• New logging keys for LLM training benchmark

  #292 opened 2 years ago by sgpyc
  2

• Remove weight init checking from compliance checker

  #299 opened 2 years ago by matthew-frank
  0

• RCP checker breaks if no RCP for batch size

  #307 opened 2 years ago by maanug-nv
  0

• Full CSV generation for MLPerf submission

  #282 opened 2 years ago by pgmpablo157321
  0

• Score norm factor from RCP checker should be exported

  #288 opened 2 years ago by maanug-nv
  0

• RCP checker finds benchmark name in surprising, undocumented way

  #254 opened 2 years ago by matthew-frank
  0

• RCP Checker functions used in other modules should be public

  #287 opened 2 years ago by maanug-nv
  0

• RCP Checker __main__.py inconsistent with other modules

  #286 opened 2 years ago by maanug-nv
  0

• Description errors for variable that changed names

  #291 opened 2 years ago by pgmpablo157321
  0

• Bert checking fails on missing "opt_epsilon" - bug or feature?

  #277 opened 2 years ago by plvlnisse
  2

• Training: add rule set for 2.1

  #256 opened 2 years ago by xyhuang
  0

• Automate results spreadsheet generation for Training v2.1

  #258 opened 2 years ago by erichan1
  9

• package_checker creates file scaling.json in result directory

  #278 opened 2 years ago by plvlnisse
  2

• Normalize score for submissions faster than RCP

  #272 opened 2 years ago by maanug-nv
  0

• No tag for Training v2.1

  #268 opened 2 years ago by mmarcinkiewicz
  0

• No hpc-2.0.0-rc1 tag

  #260 opened 2 years ago by mmarcinkiewicz
  6

• Update HPC RCPs

  #265 opened 2 years ago by maanug-nv
  0

• Handle cases where it takes too long to process logs.

  #242 opened 2 years ago by emizan76
  1

• For loop in _prune_rcps not supporting HPC

  #251 opened 2 years ago by michal2409
  0

• Incompatibility with python versions < 3.7

  #232 opened 2 years ago by eyal-lavee
  2

• Maskrcnn num_image_candidates compliance

  #245 opened 2 years ago by nv-rborkar
  0

• Result summarizer may produce broken details link

  #246 opened 2 years ago by matthew-frank
  0

• Maintain log spec

  #244 opened 2 years ago by emizan76
  0

• Check the existence of source code

  #243 opened 2 years ago by emizan76
  0

• Logs submission_* fields should match system description files

  #241 opened 2 years ago by emizan76
  0

• Finding log file

  #240 opened 2 years ago by mahmoodn
  0

• Results summarizer fails when results folder is empty

  #236 opened 3 years ago by pgmpablo157321
  0

• Set requirement for SW packages needed to run logging infra

  #235 opened 3 years ago by emizan76
  0

• Add repository checking support and include in the verify script

  #224 opened 3 years ago by emizan76
  1

• PyPI package for mllogger

  #218 opened 3 years ago by hanlint
  6

• Pass checker parameters in config file

  #205 opened 3 years ago by emizan76
  1

• Relax RCP checking rules

  #206 opened 3 years ago by emizan76
  3

• Updating repo for training v2.0 submission

  #207 opened 3 years ago by shangw-nvidia
  1

• Parameter rcp_bypass should cover case where RCP is missing

  #213 opened 3 years ago by emizan76
  1

• Use original RCP for RCP test if batch size and ALL hparams of submission match the RCP, otherwise used pruned RCPs.

  #214 opened 3 years ago by emizan76
  1

• Allow applying a scaling-factor to results for automated RCP normalization

  #210 opened 3 years ago by nv-rborkar
  2

• Modify compile time for open submissions

  #216 opened 3 years ago by emizan76
  0

• Enable weight initialization

  #212 opened 3 years ago by emizan76
  0