add oss flash fmha and fmhca support #49438
Merged
Conversation
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
|
你的PR提交成功，感谢你对开源项目的贡献! |
wwbitejotunn
force-pushed
the
sd-oss-fmha-fmhca
branch
from
4cecf1d
to
9c1e126
Compare
MARD1NO
reviewed
Jan 3, 2023
|
|
||
| void TrtCrossMultiHeadMatmulFusePass::ApplyImpl(Graph* graph) const { | ||
| FusePassBase::Init(name_scope_, graph); | ||
| #ifdef PADDLE_WITH_TENSORRT |
There was a problem hiding this comment.
我理解这里进行early stop的话是要在build fusion之前进行, 因此放在ApplyImpl的最开头.
There was a problem hiding this comment.
经过讨论后, 目前采用运行期判定的方式进行early stop, 麻烦辛苦再看一下
paddle/fluid/framework/ir/trt_flash_multihead_matmul_fuse_pass.cc
Outdated
Show resolved
Hide resolved
| FusePassBase::Init(name_scope_, graph); | ||
| auto* scope = param_scope(); | ||
|
|
||
| #ifdef PADDLE_WITH_TENSORRT |
There was a problem hiding this comment.
经过讨论后, 目前采用运行期判定的方式进行early stop, 麻烦辛苦再看一下
paddle/fluid/inference/tensorrt/convert/flash_multihead_matmul_op.cc
Outdated
Show resolved
Hide resolved
MARD1NO
previously approved these changes
Jan 6, 2023
heavengate
reviewed
Jan 6, 2023
| std::get<2>(trt_version) * 10 < | ||
| 8520) { | ||
| VLOG(3) << "Flash attention oss plugin only available for trt version >= " | ||
| "8.5.2.2. Stop this pass"; |
There was a problem hiding this comment.
看着pass有限定trt 8.5.2.2才注册，这里应该不用判断了
There was a problem hiding this comment.
考虑后还是在这里进行runtime时的early stop, 因此现在在这里加上了return
heavengate
reviewed
Jan 6, 2023
paddle/fluid/framework/ir/trt_flash_multihead_matmul_fuse_pass.cc
Outdated
Show resolved
Hide resolved
heavengate
reviewed
Jan 6, 2023
paddle/fluid/framework/ir/trt_flash_multihead_matmul_fuse_pass.cc
Outdated
Show resolved
Hide resolved
refine compile fix compile
wwbitejotunn
force-pushed
the
sd-oss-fmha-fmhca
branch
from
9691d16
to
6a679e8
Compare
heavengate
approved these changes
Jan 11, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
PR types
Performance optimization
PR changes
OPs
Describe
Add nvidia tensorrt oss plugin flash attention and cross attention support to accelerate the inference speed of stable diffusion and other models.
Using flash attention and cross attention plugin, the stable diffusion latency can be speed up from 1.52s to 1.02s.
Tensorrt 8.5.2 is required to using those plugins.
Using nsys, we can see that the plugin are successful involved by unit test under trt8.5.2.2 environment