Add new Rome model (#1304)

* Add another rome model and override

* Fix bug

* Fix typo

* Add ring

* Update ring

* Fix model matching

* Clean up

* Clean up

* Reverse rings for NCCL_RINGS input

* Only reverse NCCL_RINGS for ring graph

* Fix mapping issue when using  NCCL_RINGS

* Add NCCL_RINGS_REMAP to handle inconsistant net names

[ROCm/rccl commit: 532b70afb6]
This commit is contained in:
Wenkai Du
2024-08-23 08:45:43 +08:00
کامیت شده توسط GitHub
والد 886b576722
کامیت 157cc5f6ba
7فایلهای تغییر یافته به همراه46 افزوده شده و 10 حذف شده
@@ -21,7 +21,7 @@
DIR="$(cd -P "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
for i in {0..82}
for i in {0..87}
do
if [[ $i -eq 50 ]] || [[ $i -eq 51 ]]
then
@@ -159,6 +159,7 @@ NodeModelDesc model_descs[] = {
{2, "topo_8p_940_16n.xml", "2 nodes gfx940 16 NICs"},
{2, "topo_8p1h_6.xml", "2 nodes 8P1H Alt."},
{5, "topo_8p_940.xml", "5 nodes gfx940 8P"},
{2, "topo_8p_942.xml", "2 nodes gfx942 8P"},
};
NCCL_PARAM(MaxCTAs, "MAX_CTAS", MAXCHANNELS);