update

guided_diffusion/dist_util.py

@@ -65,7 +65,7 @@ def setup_dist(args):
 
     # dist.init_process_group(backend='nccl', init_method='env://', rank=args.local_rank, world_size=th.cuda.device_count(),  timeout=datetime.timedelta(seconds=5400))
     # st() no mark
-    dist.init_process_group(backend='nccl', init_method='env://', timeout=datetime.timedelta(seconds=54000))
+    dist.init_process_group(backend='gloo', init_method='env://', timeout=datetime.timedelta(seconds=54000))
     print(f"{args.local_rank=} init complete")
 
     # synchronize() # extra memory on rank 0, why?