github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-09-25 11:44:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix typo with colossalai/trainer utils zero (#3908)

Browse Source
This commit is contained in:
digger yu
2023-06-07 16:08:37 +08:00
committed by GitHub
parent b306cecf28
commit a9d1cadc49
15 changed files with 28 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
colossalai/zero/gemini/chunk/chunk.py
View File

@@ -416,7 +416,7 @@ class Chunk:
Copy data slice to the memory space indexed by the input tensor in the chunk.
Args:
tensor (torch.Tensor): the tensor used to retrive meta information
tensor (torch.Tensor): the tensor used to retrieve meta information
data_slice (torch.Tensor): the tensor to be copied to the chunk
"""
# sanity check

2
colossalai/zero/gemini/chunk/manager.py
View File

@@ -157,7 +157,7 @@ class ChunkManager:
Copy data to the chunk.
Args:
tensor (torch.Tensor): the tensor used to retrive meta information
tensor (torch.Tensor): the tensor used to retrieve meta information
data (torch.Tensor): the tensor to be copied to the chunk
"""
chunk = self.tensor_chunk_map[tensor]

2
colossalai/zero/gemini/memory_tracer/chunk_memstats_collector.py
View File

@@ -25,7 +25,7 @@ class ChunkMemStatsCollector(MemStatsCollector):
# override
def record_model_data_volume(self) -> None:
"""
record model data volumn on cuda and cpu.
record model data volume on cuda and cpu.
"""
if self._start_flag and not self.use_outside_memstats:
cuda_mem = self._chunk_manager.total_mem['cuda']

4
colossalai/zero/gemini/memory_tracer/memory_monitor.py
View File

@@ -45,7 +45,7 @@ class MemoryMonitor:
class AsyncMemoryMonitor(MemoryMonitor):
"""
An Async Memory Monitor runing during computing. Sampling memory usage of the current GPU
An Async Memory Monitor running during computing. Sampling memory usage of the current GPU
at interval of `1/(10**power)` sec.
The idea comes from Runtime Memory Tracer of PatrickStar
@@ -67,7 +67,7 @@ class AsyncMemoryMonitor(MemoryMonitor):
async_mem_monitor.save('log.pkl')
Args:
power (int, optional): the power of time interva. Defaults to 10.
power (int, optional): the power of time interval. Defaults to 10.
.. _PatrickStar: Parallel Training of Pre-trained Models via Chunk-based Memory Management:
https://arxiv.org/abs/2108.05818

2
colossalai/zero/gemini/utils.py
View File

@@ -73,7 +73,7 @@ def get_static_torch_model(zero_ddp_model,
zero_ddp_model (ZeroDDP): a zero ddp model
device (torch.device): the device of the final torch model
dtype (torch.dtype): the dtype of the final torch model
only_rank_0 (bool): if True, only rank0 has the coverted torch model
only_rank_0 (bool): if True, only rank0 has the converted torch model
Returns:
torch.nn.Module: a static torch model used for saving checkpoints or numeric checks