-
Notifications
You must be signed in to change notification settings - Fork 81
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
take care of flash attention for Muse, and make sure it supports the …
…normed queries and keys with custom scale
- Loading branch information
1 parent
88e188b
commit 371899e
Showing
4 changed files
with
149 additions
and
18 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,126 @@ | ||
from functools import wraps | ||
from packaging import version | ||
from collections import namedtuple | ||
|
||
import torch | ||
from torch import nn, einsum | ||
import torch.nn.functional as F | ||
|
||
# constants | ||
|
||
AttentionConfig = namedtuple('AttentionConfig', ['enable_flash', 'enable_math', 'enable_mem_efficient']) | ||
|
||
# helpers | ||
|
||
def exists(val): | ||
return val is not None | ||
|
||
def once(fn): | ||
called = False | ||
@wraps(fn) | ||
def inner(x): | ||
nonlocal called | ||
if called: | ||
return | ||
called = True | ||
return fn(x) | ||
return inner | ||
|
||
print_once = once(print) | ||
|
||
# main class | ||
|
||
class Attend(nn.Module): | ||
def __init__( | ||
self, | ||
scale = 8, | ||
dropout = 0., | ||
flash = False | ||
): | ||
super().__init__() | ||
self.scale = scale | ||
self.dropout = dropout | ||
self.attn_dropout = nn.Dropout(dropout) | ||
|
||
self.flash = flash | ||
assert not (flash and version.parse(torch.__version__) < version.parse('2.0.0')), 'in order to use flash attention, you must be using pytorch 2.0 or above' | ||
|
||
# determine efficient attention configs for cuda and cpu | ||
|
||
self.cpu_config = AttentionConfig(True, True, True) | ||
self.cuda_config = None | ||
|
||
if not torch.cuda.is_available() or not flash: | ||
return | ||
|
||
device_properties = torch.cuda.get_device_properties(torch.device('cuda')) | ||
|
||
if device_properties.major == 8 and device_properties.minor == 0: | ||
print_once('A100 GPU detected, using flash attention if input tensor is on cuda') | ||
self.cuda_config = AttentionConfig(True, False, False) | ||
else: | ||
print_once('Non-A100 GPU detected, using math or mem efficient attention if input tensor is on cuda') | ||
self.cuda_config = AttentionConfig(False, True, True) | ||
|
||
def flash_attn(self, q, k, v, mask = None): | ||
default_scale = q.shape[-1] ** -0.5 | ||
|
||
is_cuda = q.is_cuda | ||
|
||
q, k, v = map(lambda t: t.contiguous(), (q, k, v)) | ||
|
||
# scaled_dot_product_attention does not allow for custom scale | ||
# so hack it in, to support rmsnorm-ed queries and keys | ||
|
||
rescale = self.scale / default_scale | ||
|
||
q = q * (rescale ** 0.5) | ||
k = k * (rescale ** 0.5) | ||
|
||
# Check if there is a compatible device for flash attention | ||
|
||
config = self.cuda_config if is_cuda else self.cpu_config | ||
|
||
# pytorch 2.0 flash attn: q, k, v, mask, dropout, causal, softmax_scale | ||
|
||
with torch.backends.cuda.sdp_kernel(**config._asdict()): | ||
out = F.scaled_dot_product_attention( | ||
q, k, v, | ||
attn_mask = mask, | ||
dropout_p = self.dropout if self.training else 0. | ||
) | ||
|
||
return out | ||
|
||
def forward(self, q, k, v, mask = None, force_non_flash = False): | ||
""" | ||
einstein notation | ||
b - batch | ||
h - heads | ||
n, i, j - sequence length (base sequence length, source, target) | ||
d - feature dimension | ||
""" | ||
|
||
if self.flash and not force_non_flash: | ||
return self.flash_attn(q, k, v, mask = mask) | ||
|
||
# similarity | ||
|
||
sim = einsum("b h i d, b h j d -> b h i j", q, k) * self.scale | ||
|
||
# masking | ||
|
||
if exists(mask): | ||
mask_value = -torch.finfo(sim.dtype).max | ||
sim = sim.masked_fill(~mask, mask_value) | ||
|
||
# attention | ||
|
||
attn = sim.softmax(dim = -1) | ||
attn = self.attn_dropout(attn) | ||
|
||
# aggregate values | ||
|
||
out = einsum("b h i j, b h j d -> b h i d", attn, v) | ||
|
||
return out |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters