PR #2040 Comms Benchmarks

Add v1 allreduce benchmark

Quentin-Anthony committed 4 years ago

test credentials

Quentin-Anthony committed 4 years ago

Add WIP benchmarks

Quentin-Anthony committed 4 years ago

Change single runs to automatically use max message size

Quentin-Anthony committed 4 years ago

Add pt2pt and README. Make mem_factor an arg.

Quentin-Anthony committed 4 years ago

Clean up

Quentin-Anthony committed 4 years ago

Make debug alltoall-specific

Quentin-Anthony committed 4 years ago

Delete runall.py

Quentin-Anthony committed 4 years ago

Delete all_reduce_v2.py

Quentin-Anthony committed 4 years ago

Move benchmarks

Quentin-Anthony committed 4 years ago

Merge branch 'qanthony/comms-bench' of https://github.com/microsoft/DeepSpeed into qanthony/comms-bench

Quentin-Anthony committed 4 years ago

Remove unnecessary DS imports and double headers

Quentin-Anthony committed 4 years ago

Added ds_bench, abstracted out args, error handling

Quentin-Anthony committed 4 years ago

Add ds_bench to setup and README

Quentin-Anthony committed 4 years ago

No longer package ds_bench with setup.py

Quentin-Anthony committed 4 years ago

Fix ds_bench, fix args, fix dtypes, fix bw_unit

Quentin-Anthony committed 4 years ago

Fixed allgather mem_factor

Quentin-Anthony committed 4 years ago

Merge branch 'master' into qanthony/comms-bench

Quentin-Anthony committed 4 years ago

Merge branch 'master' into qanthony/comms-bench

jeffra committed 4 years ago

DeepSpeed Comms Benchmarks #2040 Merged