mirror of https://github.com/hpcaitech/ColossalAI
aibig-modeldata-parallelismdeep-learningdistributed-computingfoundation-modelsheterogeneous-traininghpcinferencelarge-scalemodel-parallelismpipeline-parallelism
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
41 lines
1.4 KiB
41 lines
1.4 KiB
import torch |
|
|
|
import colossalai |
|
from colossalai.testing import rerun_if_address_is_in_use, spawn |
|
from colossalai.zero.legacy.gemini.tensor_utils import colo_model_data_tensor_move, colo_model_data_tensor_move_inline |
|
from colossalai.zero.legacy.sharded_param import ShardedTensor |
|
|
|
|
|
def run_tensor_move(rank, world_size, port): |
|
colossalai.launch(config={}, rank=0, world_size=world_size, host='localhost', port=port, backend='nccl') |
|
|
|
src_t = torch.ones(2, 3).cuda() |
|
tgt_t = torch.zeros(2, 3) |
|
|
|
colo_model_data_tensor_move(src_t, tgt_t) |
|
assert (torch.sum(tgt_t) == 6.0), f"{torch.sum(tgt_t.payload)} vs. 6.0" |
|
|
|
src_t = torch.ones(2, 3) |
|
tgt_t = torch.zeros(2, 3).cuda().half() |
|
colo_model_data_tensor_move(src_t, tgt_t) |
|
# the src_t has been removed |
|
assert (src_t.numel() == 0) |
|
assert (torch.sum(tgt_t) == 6.0), f"{torch.sum(tgt_t.payload)} vs. 6.0" |
|
|
|
src_t = ShardedTensor(torch.ones(2, 3)) |
|
tgt_t = ShardedTensor(torch.zeros(2, 3).cuda().half()) |
|
colo_model_data_tensor_move(src_t, tgt_t) |
|
assert (torch.sum(tgt_t.payload) == 6.0), f"{torch.sum(tgt_t.payload)} vs. 6.0" |
|
|
|
assert (tgt_t.device.type == 'cuda') |
|
colo_model_data_tensor_move_inline(tgt_t, torch.device('cpu')) |
|
assert (tgt_t.device.type == 'cpu') |
|
|
|
|
|
@rerun_if_address_is_in_use() |
|
def test_tensor_move(): |
|
spawn(run_tensor_move, 1) |
|
|
|
|
|
if __name__ == '__main__': |
|
test_tensor_move()
|
|
|