mirror of https://github.com/InternLM/InternLM
add zh docs for tf32
parent
c1e30cff2c
commit
79adccc0a5
|
@ -144,4 +144,4 @@ core.*
|
||||||
# Run
|
# Run
|
||||||
llm_ckpts
|
llm_ckpts
|
||||||
events.*
|
events.*
|
||||||
memory_trace
|
memory_trace*
|
||||||
|
|
|
@ -47,6 +47,14 @@ InternLM
|
||||||
|
|
||||||
parallel
|
parallel
|
||||||
|
|
||||||
|
TF32训练
|
||||||
|
-------------------
|
||||||
|
|
||||||
|
.. toctree::
|
||||||
|
:maxdepth: 2
|
||||||
|
|
||||||
|
tf32
|
||||||
|
|
||||||
模型备份
|
模型备份
|
||||||
--------------------
|
--------------------
|
||||||
|
|
||||||
|
|
|
@ -0,0 +1,45 @@
|
||||||
|
TF32训练
|
||||||
|
==================
|
||||||
|
|
||||||
|
InternLM支持使用TF32训练模型。TensorFloat-32(TF32)是Nvidia在Ampere架构GPU上推出的专门运用于TensorCore的一种计算格式。其与其他常用数据格式的比较如下图:
|
||||||
|
|
||||||
|
.. figure:: ../../imgs/tf32.png
|
||||||
|
:scale: 50%
|
||||||
|
:class: with-border
|
||||||
|
|
||||||
|
使用TF32的前置条件:
|
||||||
|
|
||||||
|
1. 输入数据类型为FP32,且计算为矩阵乘法及卷积相关运算,才可以使用TF32作为TensorCore的中间计算类型。
|
||||||
|
|
||||||
|
|
||||||
|
2. Ampere架构的GPU。
|
||||||
|
|
||||||
|
|
||||||
|
值得注意的是,TF32仅仅是在使用TensorCore时的一种中间计算格式,并不是一个完全的数据类型。因此,为了区分不同的精度与计算格式( ``BF16`` 、``FP16`` 、``FP32`` 、``TF32`` ),InternLM支持用户在 ``model config`` 中传入 ``torch.tf32`` 来表示想要使用TF32加速运算,本质上数据类型依旧为 ``FP32``。
|
||||||
|
|
||||||
|
.. code-block:: python
|
||||||
|
|
||||||
|
model = dict(
|
||||||
|
checkpoint=False, # The proportion of layers for activation aheckpointing, the optional value are True/False/[0-1]
|
||||||
|
num_attention_heads=NUM_ATTENTION_HEAD,
|
||||||
|
embed_split_hidden=True,
|
||||||
|
vocab_size=VOCAB_SIZE,
|
||||||
|
embed_grad_scale=1,
|
||||||
|
parallel_output=True,
|
||||||
|
hidden_size=HIDDEN_SIZE,
|
||||||
|
num_layers=NUM_LAYER,
|
||||||
|
mlp_ratio=MLP_RATIO,
|
||||||
|
apply_post_layer_norm=False,
|
||||||
|
dtype="torch.tf32", # Support: "torch.float16", "torch.half", "torch.bfloat16", "torch.float32", "torch.tf32"
|
||||||
|
norm_type="rmsnorm",
|
||||||
|
layer_norm_epsilon=1e-5,
|
||||||
|
use_flash_attn=True,
|
||||||
|
num_chunks=1, # if num_chunks > 1, interleaved pipeline scheduler is used.
|
||||||
|
)
|
||||||
|
|
||||||
|
InternLM会根据 ``model config`` 中的 ``dtype`` 字符串来判断真正的数据类型。InternLM通过设置以下变量来开启TF32训练。
|
||||||
|
|
||||||
|
.. code-block:: python
|
||||||
|
|
||||||
|
torch.backends.cudnn.allow_tf32 = True
|
||||||
|
torch.backends.cuda.matmul.allow_tf32 = True
|
Binary file not shown.
After Width: | Height: | Size: 47 KiB |
Loading…
Reference in New Issue