Fixing test_detection_model_trainable_backbone_layers

…torch#10)

* fix iterate_one_epoch to avoid memory overflow and empty datasets

* use rglob in bench_decoding

* use perf_counter instead of time

* remove error for empty datasets

* [llm] Init draft NVIDIA reference

* [LLM] Add exact HPs used to match NVIDIA's convergence curves

* [LLM] Add data preprocessing steps and remove dropout

* [LLM] fix eval, add ckpt load util, remove unnecessary files

* [LLM] Update data preprocessing stage in README

* Full validation and google settings

* Apply review comments

* Anmolgupt/nvidia llm reference update (pytorch#3)

* Update Nvidia LLM reference code version

Co-authored-by: Anmol Gupta <anmolg@nvidia.com>

* fixes to imports (pytorch#5)

Co-authored-by: Anmol Gupta <14880251+anmolgupt@users.noreply.github.com>

* distributed checkpoint and mlperf logger support (pytorch#6)

* readme and mllogger keywords update (pytorch#7)

Co-authored-by: Anmol Gupta <anmolg@nvidia.com>

* Update fp32_checkpoint_checksum.log

* Update README.md

* Update README.md

* Update README.md

* mlperf logger keywords update (pytorch#8)

Co-authored-by: Anmol Gupta <anmolg@nvidia.com>

* [LLM] Create framework folder

* [LLM] Update README to follow reference template

* Describe LLM checkpoint format in README (pytorch#9)

Describe LLM checkpoint format in README

* [LLM] Readme updates, small fixes

* readme update and run script eval update (pytorch#10)

Co-authored-by: Anmol Gupta <anmolg@nvidia.com>

---------

Co-authored-by: Mikołaj Błaż <mblaz@nvidia.com>
Co-authored-by: anmolgupt <14880251+anmolgupt@users.noreply.github.com>
Co-authored-by: Anmol Gupta <anmolg@nvidia.com>
Co-authored-by: mikolajblaz <mikolajblaz@users.noreply.github.com>