Add back data parallelism (#1635)
This commit is contained in:
@@ -141,7 +141,7 @@ class ModelRunner:
|
||||
self.init_attention_backend()
|
||||
|
||||
def init_torch_distributed(self):
|
||||
logger.info("Init torch distributed begin.")
|
||||
logger.info("Init torch distributed begin.")
|
||||
# Init torch distributed
|
||||
if self.device == "cuda":
|
||||
torch.cuda.set_device(self.gpu_id)
|
||||
|
||||
Reference in New Issue
Block a user