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
This commit is contained in:
Wenkai Du
2024-08-23 08:45:43 +08:00
committed by GitHub
orang tua db840f024e
melakukan 532b70afb6
7 mengubah file dengan 46 tambahan dan 10 penghapusan
+1
Melihat File
@@ -355,6 +355,7 @@ ncclResult_t ncclTopoAddNet(struct ncclXmlNode* xmlNet, struct ncclTopoSystem* s
NCCLCHECK(xmlGetAttrIntDefault(xmlNet, "gdr", &net->net.gdrSupport, 0));
NCCLCHECK(xmlGetAttrIntDefault(xmlNet, "maxconn", &net->net.maxChannels, MAXCHANNELS));
NCCLCHECK(xmlGetAttrIntDefault(xmlNet, "coll", &net->net.collSupport, 0));
NCCLCHECK(xmlGetAttrIntDefault(xmlNet, "dev", &net->net.dev, 0));
net->net.busId = busId;
ncclDebugNoWarn = 0;