mirror of https://github.com/InternLM/InternLM
add doc for fused precision
parent
cbb26d9136
commit
95e800e10b
|
@ -133,7 +133,7 @@ ZeRO1.5 的实现使用了分层分片的概念,通过配置值 ``parallel.zer
|
|||
|
||||
hybrid_zero_optimizer = dict(
|
||||
# Enable low_level_optimzer overlap_communication
|
||||
overlap_sync_grad=True,
|
||||
overlap_sync_grad=True,
|
||||
overlap_sync_param=True,
|
||||
# bucket size for nccl communication params
|
||||
reduce_bucket_size=512 * 1024 * 1024,
|
||||
|
@ -150,3 +150,40 @@ ZeRO1.5 的实现使用了分层分片的概念,通过配置值 ``parallel.zer
|
|||
|
||||
.. autoclass:: internlm.solver.optimizer.hybrid_zero_optim.HybridZeroOptimizer
|
||||
:members:
|
||||
|
||||
混合精度
|
||||
-----------------
|
||||
混合精度是指在模型训练的过程中同时使用16位和32位浮点类型,是一种在最小化精度损失的前提下加速模型训练的方法。
|
||||
混合精度通过让模型的某些部分使用32位浮点数以保持数值稳定性,并在其余部分利用半精度浮点数加速训练并减少内存使用,在评估指标(如准确率)方面仍可以获得同等的训练效果。
|
||||
|
||||
.. autoclass:: internlm.core.naive_amp.NaiveAMPModel
|
||||
|
||||
InternLM默认将模型转换为16位精度进行训练(在配置文件中可以设置默认类型为其他数据类型)。在使用混合精度时,需要在构建模型时使用
|
||||
|
||||
.. code-block:: python
|
||||
|
||||
set_fp32_attr_to_module(/*fp32 module*/)
|
||||
|
||||
将模型的某个子模块设置为32精度进行训练,InternLM会在模型训练时自动将数据类型转换成需要的精度。
|
||||
|
||||
例如:
|
||||
|
||||
.. code-block:: python
|
||||
|
||||
class MlpModel(nn.Module):
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
self.linear1 = nn.Linear(4, 1, bias=False)
|
||||
self.linear2 = nn.Linear(1, 4, bias=False)
|
||||
|
||||
model = MlpModel()
|
||||
# 将model.linear2设置为fp32模块
|
||||
set_fp32_attr_to_module(model.linear2)
|
||||
|
||||
# 混合精度模型
|
||||
model = NaiveAMPModel(
|
||||
model=model,
|
||||
output_to_fp32=True,
|
||||
dtype=torch.bfloat16(),
|
||||
sync_buffer=False,
|
||||
)
|
||||
|
|
|
@ -148,6 +148,10 @@ class NaiveAMPModel(nn.Module):
|
|||
return out
|
||||
|
||||
def _register_fp32_parameters_hook(self) -> None:
|
||||
"""
|
||||
Set module to fp32 and register automatic conversion hook in the forward pass.
|
||||
The fp32 modules are marked by set_fp32_attr_to_module(.)
|
||||
"""
|
||||
dtype = torch.float32
|
||||
|
||||
def to_fp32(x, dtype=dtype):
|
||||
|
|
Loading…
Reference in New Issue