created megatron package

model/distributed.py → megatron/model/distributed.py

@@ -19,7 +19,7 @@ import torch.distributed as dist
 from torch.nn.modules import Module
 from torch.autograd import Variable
 
-import mpu
+from megatron import mpu
 
 class DistributedDataParallel(Module):
 

model/gpt2_modeling.py → megatron/model/gpt2_modeling.py

@@ -18,7 +18,7 @@
 import torch
 import torch.nn.functional as F
 
-import mpu
+from megatron import mpu
 
 
 def init_method_normal(std=0.02):

model/modeling.py → megatron/model/modeling.py

@@ -34,9 +34,9 @@ from torch.nn import CrossEntropyLoss
 
 #from torch.utils.checkpoint import checkpoint
 
-from data_utils.file_utils import cached_path
+from megatron.data_utils.file_utils import cached_path
 
-import mpu
+from megatron import mpu
 
 
 def normal_init_method(mean, std):