Skip to content

Releases: mlcommons/storage

v1.0.1 release

07 Aug 18:10
171a418
Compare
Choose a tag to compare

Patch release for memory fix observed for resnet50 workload benchmark run with high number of accelerators.

v1.0 release

26 Jun 02:20
cd848ba
Compare
Choose a tag to compare

MLPerf Storage v1.0 benchmark release

v1.0-rc1 release

15 Mar 16:25
bb3a1d2
Compare
Choose a tag to compare
v1.0-rc1 release Pre-release
Pre-release

Second preview release for MLPerf Storage Benchmark v1.0

v1.0-rc0 release

25 Feb 17:57
edd2a28
Compare
Choose a tag to compare
v1.0-rc0 release Pre-release
Pre-release

First preview release for MLPerf Storage Benchmark v1.0

v0.5 Release

14 Jun 17:37
aca225d
Compare
Choose a tag to compare

v0.5 release of the MLPerf Storage benchmark

v0.5-rc2 release

24 May 16:45
673843a
Compare
Choose a tag to compare
v0.5-rc2 release Pre-release
Pre-release

Key features in v0.5-rc2 release

  1. Larger dataset can be generated while subset can be used for the benchmark run
  2. Extra validations added for model name, dataset size and accelerators

v0.5-rc1 release

06 Apr 16:27
bd21330
Compare
Choose a tag to compare
v0.5-rc1 release Pre-release
Pre-release

Key features in v0.5-rc1 release

  1. New command datasize to calculate the minimum dataset generated for the benchmark given number of accelerators and client host memory.
  2. Report summary.json is created in the output folder after every benchmark run. It reports accelerator utilisation, benchmark success/failure(if it meets AU requirements) , throughput in samples/sec and MB/sec
  3. New command reportgen to create accumulated report for multi host runs

v0.5-rc0 release

03 Feb 14:29
e6c0167
Compare
Choose a tag to compare
v0.5-rc0 release Pre-release
Pre-release

First preview release for MLPerf Storage Benchmark v0.5