intermediate/scaled_dot_product_attention_tutorial
在 Google Colab 中运行
Colab
下载 Notebook
Notebook
在 GitHub 上查看
GitHub
注意
跳转到末尾 下载完整的示例代码。
(Beta) 使用缩放点积注意力（SDPA）实现高性能 Transformer#
创建日期: 2023年3月15日 | 最后更新: 2024年10月09日 | 最后验证: 2024年11月05日
作者: Driss Guessous
摘要#
在本教程中，我们想重点介绍一个有助于实现 Transformer 架构的新 torch.nn.functional 函数。该函数名为 torch.nn.functional.scaled_dot_product_attention。有关该函数的详细描述，请参阅 PyTorch 文档。该函数已集成到 torch.nn.MultiheadAttention 和 torch.nn.TransformerEncoderLayer 中。
概述#
总的来说，此 PyTorch 函数根据论文 Attention is all you need 中的定义，计算查询（query）、键（key）和值（value）之间的缩放点积注意力（SDPA）。虽然可以使用现有的 PyTorch 函数来实现此功能，但融合实现可以比朴素实现带来显著的性能优势。
融合实现#
对于 CUDA 张量输入，该函数将分派到以下实现之一：
用 C++ 定义的 PyTorch 实现
注意
本教程需要 PyTorch 2.0.0 或更高版本。
import torch
import torch.nn as nn
import torch.nn.functional as F
device = "cuda" if torch.cuda.is_available() else "cpu"
# Example Usage:
query, key, value = torch.randn(2, 3, 8, device=device), torch.randn(2, 3, 8, device=device), torch.randn(2, 3, 8, device=device)
F.scaled_dot_product_attention(query, key, value)
tensor([[[-0.1471, 0.0784, -0.0581, -0.5448, 0.0610, -0.4824, 0.0488,
-0.4969],
[-0.3822, -0.5073, -0.2710, -0.7289, 0.1801, -0.2160, -0.0845,
-0.1191],
[-0.3038, -0.4056, -0.3013, -0.4887, 0.2677, -0.2204, -0.0220,
-0.0686]],
[[ 0.1158, -0.4914, 1.2867, -0.2343, 0.2195, -0.3615, 0.2703,
-1.0827],
[ 0.1269, -0.4285, 1.2088, -0.3356, 0.1363, -0.2540, 0.3196,
-0.9992],
[ 0.1568, -0.4041, 1.2502, -0.3362, 0.1297, -0.2964, 0.3251,
-0.9878]]], device='cuda:0')
显式分派控制#
虽然函数会自动分派到三种实现之一，但用户也可以通过使用上下文管理器来显式控制分派。此上下文管理器允许用户显式禁用某些实现。如果用户想确保函数确实使用了最快的实现来处理其特定输入，可以使用上下文管理器来遍历并测量性能。
# Lets define a helpful benchmarking function:
import torch.utils.benchmark as benchmark
def benchmark_torch_function_in_microseconds(f, *args, **kwargs):
t0 = benchmark.Timer(
stmt="f(*args, **kwargs)", globals={"args": args, "kwargs": kwargs, "f": f}
)
return t0.blocked_autorange().mean * 1e6
# Lets define the hyper-parameters of our input
batch_size = 32
max_sequence_len = 1024
num_heads = 32
embed_dimension = 32
dtype = torch.float16
query = torch.rand(batch_size, num_heads, max_sequence_len, embed_dimension, device=device, dtype=dtype)
key = torch.rand(batch_size, num_heads, max_sequence_len, embed_dimension, device=device, dtype=dtype)
value = torch.rand(batch_size, num_heads, max_sequence_len, embed_dimension, device=device, dtype=dtype)
print(f"The default implementation runs in {benchmark_torch_function_in_microseconds(F.scaled_dot_product_attention, query, key, value):.3f} microseconds")
# Lets explore the speed of each of the 3 implementations
from torch.nn.attention import SDPBackend, sdpa_kernel
with sdpa_kernel(SDPBackend.MATH):
math_time=benchmark_torch_function_in_microseconds(F.scaled_dot_product_attention, query, key, value)
print(f"The math implementation runs in {math_time:.3f} microseconds")
with sdpa_kernel(SDPBackend.FLASH_ATTENTION):
try:
flash_time=benchmark_torch_function_in_microseconds(F.scaled_dot_product_attention, query, key, value)
print(f"The flash attention implementation runs in {flash_time:.3f} microseconds")
except RuntimeError:
print("FlashAttention is not supported. See warnings for reasons.")
with sdpa_kernel(SDPBackend.EFFICIENT_ATTENTION):
try:
efficient_time=benchmark_torch_function_in_microseconds(F.scaled_dot_product_attention, query, key, value)
print(f"The memory efficient implementation runs in {efficient_time:.3f} microseconds")
except RuntimeError:
print("EfficientAttention is not supported. See warnings for reasons.")
The default implementation runs in 2273.824 microseconds
The math implementation runs in 87451.818 microseconds
The flash attention implementation runs in 2281.270 microseconds
The memory efficient implementation runs in 4357.890 microseconds
硬件依赖性#
根据您运行上述单元格的机器以及可用的硬件，您的结果可能会有所不同。 - 如果您没有 GPU 并且在 CPU 上运行，那么对于 FP32，上下文管理器将不起作用，所有三次运行都应返回相似的时间。 - 根据您的显卡支持的计算能力，Flash Attention 或内存高效实现可能已失败。
因果自注意力#
下面是一个多头因果自注意力块的示例实现，灵感来自 Andrej Karpathy 的 NanoGPT 仓库。
class CausalSelfAttention(nn.Module):
def __init__(self, num_heads: int, embed_dimension: int, bias: bool=False, is_causal: bool=False, dropout:float=0.0):
super().__init__()
assert embed_dimension % num_heads == 0
# key, query, value projections for all heads, but in a batch
self.c_attn = nn.Linear(embed_dimension, 3 * embed_dimension, bias=bias)
# output projection
self.c_proj = nn.Linear(embed_dimension, embed_dimension, bias=bias)
# regularization
self.dropout = dropout
self.resid_dropout = nn.Dropout(dropout)
self.num_heads = num_heads
self.embed_dimension = embed_dimension
# Perform causal masking
self.is_causal = is_causal
def forward(self, x):
# calculate query, key, values for all heads in batch and move head forward to be the batch dim
query_projected = self.c_attn(x)
batch_size = query_projected.size(0)
embed_dim = query_projected.size(2)
head_dim = embed_dim // (self.num_heads * 3)
query, key, value = query_projected.chunk(3, -1)
query = query.view(batch_size, -1, self.num_heads, head_dim).transpose(1, 2)
key = key.view(batch_size, -1, self.num_heads, head_dim).transpose(1, 2)
value = value.view(batch_size, -1, self.num_heads, head_dim).transpose(1, 2)
if self.training:
dropout = self.dropout
is_causal = self.is_causal
else:
dropout = 0.0
is_causal = False
y = F.scaled_dot_product_attention(query, key, value, attn_mask=None, dropout_p=dropout, is_causal=is_causal)
y = y.transpose(1, 2).view(batch_size, -1, self.num_heads * head_dim)
y = self.resid_dropout(self.c_proj(y))
return y
num_heads = 8
heads_per_dim = 64
embed_dimension = num_heads * heads_per_dim
dtype = torch.float16
model = CausalSelfAttention(num_heads=num_heads, embed_dimension=embed_dimension, bias=False, is_causal=True, dropout=0.1).to("cuda").to(dtype).eval()
print(model)
CausalSelfAttention(
(c_attn): Linear(in_features=512, out_features=1536, bias=False)
(c_proj): Linear(in_features=512, out_features=512, bias=False)
(resid_dropout): Dropout(p=0.1, inplace=False)
)
NestedTensor 和密集张量支持#
SDPA 支持 NestedTensor 和密集张量输入。 NestedTensors 可以处理输入是可变长度序列批次的情况，而无需将每个序列填充到批次中的最大长度。有关 NestedTensors 的更多信息，请参阅 torch.nested 和 NestedTensors 教程。
import random
def generate_rand_batch(
batch_size,
max_sequence_len,
embed_dimension,
pad_percentage=None,
dtype=torch.float16,
device="cuda",
):
if not pad_percentage:
return (
torch.randn(
batch_size,
max_sequence_len,
embed_dimension,
dtype=dtype,
device=device,
),
None,
)
# Random sequence lengths
seq_len_list = [
int(max_sequence_len * (1 - random.gauss(pad_percentage, 0.01)))
for _ in range(batch_size)
]
# Make random entry in the batch have max sequence length
seq_len_list[random.randint(0, batch_size - 1)] = max_sequence_len
return (
torch.nested.nested_tensor(
[
torch.randn(seq_len, embed_dimension,
dtype=dtype, device=device)
for seq_len in seq_len_list
]
),
seq_len_list,
)
random_nt, _ = generate_rand_batch(32, 512, embed_dimension, pad_percentage=0.5, dtype=dtype, device=device)
random_dense, _ = generate_rand_batch(32, 512, embed_dimension, pad_percentage=None, dtype=dtype, device=device)
# Currently the fused implementations don't support ``NestedTensor`` for training
model.eval()
with sdpa_kernel(SDPBackend.FLASH_ATTENTION):
try:
print(f"Random NT runs in {benchmark_torch_function_in_microseconds(model, random_nt):.3f} microseconds")
print(f"Random Dense runs in {benchmark_torch_function_in_microseconds(model, random_dense):.3f} microseconds")
except RuntimeError:
print("FlashAttention is not supported. See warnings for reasons.")
/usr/local/lib/python3.10/dist-packages/torch/nested/__init__.py:250: UserWarning:
The PyTorch API of nested tensors is in prototype stage and will change in the near future. We recommend specifying layout=torch.jagged when constructing a nested tensor, as this layout receives active development, has better operator coverage, and works with torch.compile. (Triggered internally at /pytorch/aten/src/ATen/NestedTensorImpl.cpp:178.)
Random NT runs in 606.955 microseconds
Random Dense runs in 952.497 microseconds
将 SDPA 与 torch.compile 一起使用#
随着 PyTorch 2.0 的发布，引入了一个名为 torch.compile() 的新功能，它可以提供比即时模式显着更好的性能。缩放点积注意力与 torch.compile() 完全可组合。为了演示这一点，让我们使用 torch.compile() 编译 CausalSelfAttention 模块，并观察由此产生的性能改进。
batch_size = 32
max_sequence_len = 256
x = torch.rand(batch_size, max_sequence_len,
embed_dimension, device=device, dtype=dtype)
print(
f"The non compiled module runs in {benchmark_torch_function_in_microseconds(model, x):.3f} microseconds")
compiled_model = torch.compile(model)
# Let's compile it
compiled_model(x)
print(
f"The compiled module runs in {benchmark_torch_function_in_microseconds(compiled_model, x):.3f} microseconds")
The non compiled module runs in 425.073 microseconds
The compiled module runs in 544.064 microseconds
确切的执行时间取决于机器，但我的结果是：未编译的模块运行时间为 166.616 微秒，编译后的模块运行时间为 166.726 微秒。这并非我们所期望的。让我们深入研究一下。PyTorch 配备了一个出色的内置分析器，您可以使用它来检查代码的性能特征。
from torch.profiler import profile, record_function, ProfilerActivity
activities = [ProfilerActivity.CPU]
if device == 'cuda':
activities.append(ProfilerActivity.CUDA)
with profile(activities=activities, record_shapes=False) as prof:
with record_function(" Non-Compilied Causal Attention"):
for _ in range(25):
model(x)
print(prof.key_averages().table(sort_by="cuda_time_total", row_limit=10))
with profile(activities=activities, record_shapes=False) as prof:
with record_function("Compiled Causal Attention"):
for _ in range(25):
compiled_model(x)
print(prof.key_averages().table(sort_by="cuda_time_total", row_limit=10))
# For even more insights, you can export the trace and use ``chrome://tracing`` to view the results
#
# .. code-block:: python
#
# prof.export_chrome_trace("compiled_causal_attention_trace.json").
------------------------------------------------------- ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------
Name Self CPU % Self CPU CPU total % CPU total CPU time avg Self CUDA Self CUDA % CUDA total CUDA time avg # of Calls
------------------------------------------------------- ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------
Non-Compilied Causal Attention 16.90% 2.143ms 76.94% 9.754ms 9.754ms 0.000us 0.00% 10.836ms 10.836ms 1
Non-Compilied Causal Attention 0.00% 0.000us 0.00% 0.000us 0.000us 10.734ms 101.14% 10.734ms 10.734ms 1
aten::linear 1.05% 133.531us 35.29% 4.474ms 89.487us 0.000us 0.00% 8.012ms 160.232us 50
aten::matmul 2.01% 254.573us 31.58% 4.003ms 80.066us 0.000us 0.00% 8.012ms 160.232us 50
aten::mm 9.80% 1.242ms 27.24% 3.453ms 69.057us 7.789ms 73.39% 8.012ms 160.232us 50
ampere_fp16_s1688gemm_fp16_128x128_ldg8_f2f_tn 0.00% 0.000us 0.00% 0.000us 0.000us 5.572ms 52.51% 5.572ms 222.894us 25
aten::scaled_dot_product_attention 1.63% 207.093us 15.32% 1.943ms 77.701us 0.000us 0.00% 2.824ms 112.966us 25
aten::_scaled_dot_product_flash_attention 2.35% 298.135us 13.69% 1.735ms 69.417us 0.000us 0.00% 2.824ms 112.966us 25
aten::_flash_attention_forward 2.38% 301.874us 9.51% 1.206ms 48.225us 2.824ms 26.61% 2.824ms 112.966us 25
void pytorch_flash::flash_fwd_kernel<Flash_fwd_kerne... 0.00% 0.000us 0.00% 0.000us 0.000us 2.824ms 26.61% 2.824ms 112.966us 25
------------------------------------------------------- ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------
Self CPU time total: 12.678ms
Self CUDA time total: 10.613ms
------------------------------------------------------- ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------
Name Self CPU % Self CPU CPU total % CPU total CPU time avg Self CUDA Self CUDA % CUDA total CUDA time avg # of Calls
------------------------------------------------------- ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------
## Call CompiledFxGraph fi6oafvta3xbcesbp2mjppjhzijz... 0.00% 0.000us 0.00% 0.000us 0.000us 10.648ms 100.38% 10.648ms 425.919us 25
Compiled Causal Attention 7.05% 917.762us 86.03% 11.205ms 11.205ms 0.000us 0.00% 10.608ms 10.608ms 1
Torch-Compiled Region: 0/0 7.21% 938.613us 75.92% 9.888ms 395.521us 0.000us 0.00% 10.608ms 424.325us 25
CompiledFunction 8.70% 1.133ms 66.38% 8.645ms 345.804us 0.000us 0.00% 10.608ms 424.325us 25
## Call CompiledFxGraph fi6oafvta3xbcesbp2mjppjhzijz... 19.62% 2.555ms 57.68% 7.513ms 300.503us 0.000us 0.00% 10.608ms 424.325us 25
aten::mm 7.55% 982.923us 11.56% 1.506ms 30.115us 7.786ms 73.39% 7.786ms 155.715us 50
ampere_fp16_s1688gemm_fp16_128x128_ldg8_f2f_tn 0.00% 0.000us 0.00% 0.000us 0.000us 5.570ms 52.51% 5.570ms 222.808us 25
aten::_scaled_dot_product_flash_attention 1.79% 232.953us 12.95% 1.686ms 67.447us 0.000us 0.00% 2.822ms 112.896us 25
aten::_flash_attention_forward 2.40% 313.225us 8.96% 1.166ms 46.660us 2.822ms 26.61% 2.822ms 112.896us 25
void pytorch_flash::flash_fwd_kernel<Flash_fwd_kerne... 0.00% 0.000us 0.00% 0.000us 0.000us 2.822ms 26.61% 2.822ms 112.896us 25
------------------------------------------------------- ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------
Self CPU time total: 13.024ms
Self CUDA time total: 10.608ms
前面的代码片段生成了一个报告，显示了编译和未编译模块在 GPU 上消耗最多执行时间的 Top 10 PyTorch 函数。分析表明，对于两个模块，大部分 GPU 时间都集中在相同的函数集上。此处的原因是 torch.compile 非常擅长消除与 PyTorch 相关的框架开销。如果您的模型启动了大型、高效的 CUDA 内核（在本例中 CausalSelfAttention 就是如此），那么 PyTorch 的开销可能就会被隐藏。
实际上，您的模块通常不只包含一个 CausalSelfAttention 块。在试验 Andrej Karpathy 的 NanoGPT 仓库时，将模块编译后，每次训练步骤的时间从 6090.49ms 减少到 3273.17ms！这是在 NanoGPT 使用 Shakespeare 数据集进行训练的提交 ae3a8d5 上完成的。
将 SDPA 与 attn_bias 子类一起使用#
# As of PyTorch 2.3, we have added a new submodule that contains tensor subclasses.
# Designed to be used with ``torch.nn.functional.scaled_dot_product_attention``.
# The module is named ``torch.nn.attention.bias`` and contains the following two
# utilities for generating causal attention variants:
#
# - ``torch.nn.attention.bias.causal_upper_left``
# - ``torch.nn.attention.bias.causal_lower_right``
#
# .. note::
# The current argument ``is_causal`` in ``torch.nn.functional.scaled_dot_product_attention``
# is the same as using ``torch.nn.attention.bias.causal_upper_left``.
#
from torch.nn.attention.bias import causal_lower_right, causal_upper_left
batch_size = 32
sequence_length_q = 2
sequence_length_kv = 10
num_heads = 16
embed_dimension = 32
dtype = torch.float16
query = torch.rand(batch_size, num_heads, sequence_length_q, embed_dimension, device=device, dtype=dtype)
key = torch.rand(batch_size, num_heads, sequence_length_kv, embed_dimension, device=device, dtype=dtype)
value = torch.rand(batch_size, num_heads, sequence_length_kv, embed_dimension, device=device, dtype=dtype)
upper_left_bias = causal_upper_left(sequence_length_q, sequence_length_kv)
lower_right_bias = causal_lower_right(sequence_length_q, sequence_length_kv)
print(type(upper_left_bias))
print(type(lower_right_bias))
assert type(upper_left_bias) == type(lower_right_bias)
assert issubclass(type(upper_left_bias), torch.Tensor)
# As you can see from the previous output, are the same type ``torch.nn.attention.bias.CausalBias``
# and subclass ``torch.Tensor``
# Lets see what these tensors look like
print(upper_left_bias)
print(lower_right_bias)
# Upper Left Bias aligns the causal attention mask to the upper left corner of the attention scores matrix.
# This only has an impact when the attention scores matrix is not square, which is common for decoding use cases.
# Another way of thinking about this concept is that when you use upper left bias,
# the 0th token in the query is aligned to the 0th token in the key, while for lower right bias,
# Assuming the attention score matrix is two dimensional, ``attn_score[0][0]`` is the attention score
# between the 0th token in the query and the 0th token in the key.
# For lower right bias, the sequence of q is aligned so that the last token in q is aligned to the last token in k
# (for example, ``attn_score[-1][-1])`` is all True since the last token in q is at the same position as the last token in k
# even if the sequence length of q and k are different.
# These objects are intended to be used with sdpa
out_upper_left = F.scaled_dot_product_attention(query, key, value, upper_left_bias)
out_lower_right = F.scaled_dot_product_attention(query, key, value, lower_right_bias)
out_is_causal = F.scaled_dot_product_attention(query, key, value, is_causal=True)
assert torch.allclose(out_upper_left, out_is_causal)
assert not torch.allclose(out_upper_left, out_lower_right)
# These attention biases should also be compatible with torch.compile
compiled_sdpa = torch.compile(F.scaled_dot_product_attention, fullgraph=True)
out_upper_left = compiled_sdpa(query, key, value, upper_left_bias)
<class 'torch.nn.attention.bias.CausalBias'>
<class 'torch.nn.attention.bias.CausalBias'>
tensor([[ True, False, False, False, False, False, False, False, False, False],
[ True, True, False, False, False, False, False, False, False, False]])
tensor([[ True, True, True, True, True, True, True, True, True, False],
[ True, True, True, True, True, True, True, True, True, True]])
结论#
在本教程中，我们演示了 torch.nn.functional.scaled_dot_product_attention 的基本用法。我们展示了如何使用 sdpa_kernel 上下文管理器来断言在 GPU 上使用了某种实现。此外，我们构建了一个简单的 CausalSelfAttention 模块，它可以与 NestedTensor 一起工作并且可以进行 torch 编译。在此过程中，我们展示了如何使用分析工具来探索用户定义的模块的性能特征。
脚本总运行时间: (0 分 7.096 秒)
