mirror of https://github.com/InternLM/InternLM
add docs for tf32 in mix precision
yingtongxiong
parent
27930174ae
commit
e57a99a810
|
|
@ -8,7 +8,7 @@ msgid ""
|
|||
msgstr ""
|
||||
"Project-Id-Version: InternLM \n"
|
||||
"Report-Msgid-Bugs-To: \n"
|
||||
"POT-Creation-Date: 2023-09-26 17:04+0800\n"
|
||||
"POT-Creation-Date: 2023-09-27 10:59+0800\n"
|
||||
"PO-Revision-Date: YEAR-MO-DA HO:MI+ZONE\n"
|
||||
"Last-Translator: FULL NAME <EMAIL@ADDRESS>\n"
|
||||
"Language: en\n"
|
||||
|
|
@ -83,3 +83,35 @@ msgstr ""
|
|||
#: ../../source/mixed_precision.rst:16
|
||||
msgid "例如："
|
||||
msgstr "For example:"
|
||||
|
||||
#: ../../source/mixed_precision.rst:40
|
||||
msgid "TF32训练"
|
||||
msgstr ""
|
||||
|
||||
#: ../../source/mixed_precision.rst:41
|
||||
msgid "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："
|
||||
msgstr "TensorFloat-32 (TF32) is a computational format introduced by Nvidia on Ampere Architecture GPUs for TensorCore. A comparison with other data formats is shown below."
|
||||
|
||||
#: ../../source/mixed_precision.rst:47
|
||||
msgid "使用TF32的前置条件："
|
||||
msgstr "Prerequisites for using TF32."
|
||||
|
||||
#: ../../source/mixed_precision.rst:49
|
||||
msgid "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。"
|
||||
msgstr "The input data type should be FP32 and TF32 is designed for matrix multiplication, convolutions, and other relative computations."
|
||||
|
||||
#: ../../source/mixed_precision.rst:51
|
||||
msgid "Ampere架构的GPU。"
|
||||
msgstr "Ampere Architecture GPU"
|
||||
|
||||
#: ../../source/mixed_precision.rst:53
|
||||
msgid "InternLM支持使用TF32训练模型，允许用户在config文件中将 ``dtype`` 设置为 ``torch.tf32``。"
|
||||
msgstr "InternLM supports training model in TF32 and allows user to set the ``dtype`` in config as ``torch.tf32``."
|
||||
|
||||
#: ../../source/mixed_precision.rst:75
|
||||
msgid ""
|
||||
"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，在InternLM中，尽管用户将 "
|
||||
"``dtype`` 设置成了 ``torch.tf32``，模型的数据类型依旧是 ``torch.float32``。InternLM会针对 "
|
||||
"``dtype`` 为 ``torch.tf32`` 的情况，设置以下变量来开启TF32训练。"
|
||||
msgstr "It is noticed that TF32 is an intermediate format in TensorCore instead of a data type. Therefore, InternLM could set the following environment variables to enable TF32 when the ``dtype`` is ``torch.tf32``, which is actually ``torch.float32``."
|
||||
|
||||
|
|
|
|||
|
|
@ -8,7 +8,7 @@ msgid ""
|
|||
msgstr ""
|
||||
"Project-Id-Version: InternLM \n"
|
||||
"Report-Msgid-Bugs-To: \n"
|
||||
"POT-Creation-Date: 2023-09-11 14:25+0800\n"
|
||||
"POT-Creation-Date: 2023-09-27 10:59+0800\n"
|
||||
"PO-Revision-Date: YEAR-MO-DA HO:MI+ZONE\n"
|
||||
"Last-Translator: FULL NAME <EMAIL@ADDRESS>\n"
|
||||
"Language: en\n"
|
||||
|
|
@ -360,6 +360,28 @@ msgstr ""
|
|||
"Taking the configuration of the demo training on a single machine with 8 "
|
||||
"GPUs on slurm as an example, the training result log is shown below:"
|
||||
|
||||
#: ../../../usage.md:373
|
||||
msgid "长文本生成"
|
||||
msgstr ""
|
||||
|
||||
#: ../../../usage.md:375
|
||||
msgid ""
|
||||
"在推理阶段，您可以在模型配置中通过设置 `use_dynamic_ntk_rope=True` 开启 RoPE 的 Dynamic NTK "
|
||||
"选项，从而使得模型适应长文本输入输出，达到 16K 的外推效果:"
|
||||
msgstr ""
|
||||
|
||||
#: ../../../usage.md:401
|
||||
msgid "关于 Dyanmic NTK 的原理，详细请参考"
|
||||
msgstr ""
|
||||
|
||||
#: ../../../usage.md:403
|
||||
msgid "https://www.reddit.com/r/LocalLLaMA/comments/14mrgpr/dynamically_scaled_rope_further_increases"
|
||||
msgstr ""
|
||||
|
||||
#: ../../../usage.md:404
|
||||
msgid "https://kexue.fm/archives/9675"
|
||||
msgstr ""
|
||||
|
||||
#~ msgid "`load_model_only_folder`与`load_ckpt_folder`不能同时设置"
|
||||
#~ msgstr ""
|
||||
#~ "`load_model_only_folder` and `load_ckpt_folder` "
|
||||
|
|
|
|||
|
|
@ -34,3 +34,48 @@ InternLM默认将模型转换为16位浮点数类型进行训练（在配置文
|
|||
dtype=torch.bfloat16(),
|
||||
sync_buffer=False,
|
||||
)
|
||||
|
||||
|
||||
TF32训练
|
||||
-----------------
|
||||
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。
|
||||
|
||||
InternLM支持使用TF32训练模型，允许用户在config文件中将 ``dtype`` 设置为 ``torch.tf32``。
|
||||
|
||||
.. 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.
|
||||
)
|
||||
|
||||
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，在InternLM中，尽管用户将 ``dtype`` 设置成了 ``torch.tf32``，模型的数据类型依旧是 ``torch.float32``。InternLM会针对 ``dtype`` 为 ``torch.tf32`` 的情况，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
|
||||
|
||||
|
|
|
|||
|
|
@ -1,52 +0,0 @@
|
|||
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：
|
||||
|
||||
InternLM supports training models using TF32. TensorFloat-32 (TF32) is a computational format introduced by Nvidia for TensorCores on Ampere architecture GPUs. Here's a comparison of TF32 with other data formats:
|
||||
|
||||
.. figure:: ../../imgs/tf32.png
|
||||
:scale: 50%
|
||||
:class: with-border
|
||||
|
||||
使用TF32的前置条件：
|
||||
|
||||
Prerequisites for using TF32:
|
||||
input data must be of type FP32 (single-precision floating-point) and the computations should be matrix multiplication, convolution and so on.
|
||||
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。
|
||||
|
||||
Ampere GPU
|
||||
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``。
|
||||
|
||||
It is noticed that TF32 is an intermediate calculation format when employing TensorCores. InternLM allows users to speficy ``torch.tf32`` in the model config to using TF32 acceleration while dtype is still ``torch.float32``.
|
||||
|
||||
.. 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。
|
||||
InternLM enables TF32 training by setting the following variables.
|
||||
|
||||
.. code-block:: python
|
||||
|
||||
torch.backends.cudnn.allow_tf32 = True
|
||||
torch.backends.cuda.matmul.allow_tf32 = True
|
||||
Loading…
Reference in New Issue