github
/
ColossalAI
mirror of https://github.com/hpcaitech/ColossalAI
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

[doc] updated kernel-related optimisers' docstring (#2385) 

* [doc] updated kernel-related optimisers' docstring

* polish doc
pull/2410/head
Frank Lee 2023-01-09 17:13:53 +08:00 committed by GitHub
parent 85e045b063
commit 551cafec14
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
5 changed files with 5 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
colossalai/nn/optimizer/cpu_adam.py
View File

@ -19,7 +19,7 @@ class CPUAdam(NVMeOptimizer):
* Parameters on GPU and gradients on GPU is allowed.
* Parameters on GPU and gradients on CPU is **not** allowed.
Requires ColossalAI to be installed via ``pip install .``.
`CPUAdam` requires CUDA extensions which can be built during installation or runtime.
This version of CPU Adam accelates parameters updating on CPU with SIMD.
Support of AVX2 or AVX512 is required.

3
colossalai/nn/optimizer/fused_adam.py
View File

@ -9,8 +9,7 @@ from colossalai.utils import multi_tensor_applier
class FusedAdam(torch.optim.Optimizer):
"""Implements Adam algorithm.
Currently GPU-only. Requires ColossalAI to be installed via
``pip install .``.
`FusedAdam` requires CUDA extensions which can be built during installation or runtime.
This version of fused Adam implements 2 fusions.

3
colossalai/nn/optimizer/fused_lamb.py
View File

@ -9,8 +9,7 @@ from colossalai.utils import multi_tensor_applier
class FusedLAMB(torch.optim.Optimizer):
"""Implements LAMB algorithm.
Currently GPU-only. Requires ColossalAI to be installed via
``pip install .``.
`FusedLAMB` requires CUDA extensions which can be built during installation or runtime.
This version of fused LAMB implements 2 fusions.

3
colossalai/nn/optimizer/fused_sgd.py
View File

@ -10,8 +10,7 @@ from colossalai.utils import multi_tensor_applier
class FusedSGD(Optimizer):
r"""Implements stochastic gradient descent (optionally with momentum).
Currently GPU-only. Requires ColossalAI to be installed via
``pip install .``.
`FusedSGD` requires CUDA extensions which can be built during installation or runtime.
This version of fused SGD implements 2 fusions.

2
colossalai/nn/optimizer/hybrid_adam.py
View File

@ -19,7 +19,7 @@ class HybridAdam(NVMeOptimizer):
* Parameters on GPU and gradients on GPU is allowed.
* Parameters on GPU and gradients on CPU is **not** allowed.
Requires ColossalAI to be installed via ``pip install .``
`HybriadAdam` requires CUDA extensions which can be built during installation or runtime.
This version of Hybrid Adam is an hybrid of CPUAdam and FusedAdam.