[auto parallel] complete elementwise ops spmd rules of LLaMa2 for eager semi auto parallel #58474
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
cxxly
changed the title
cxxly
changed the title
cxxly
force-pushed
the
spmd_elementwise
branch
3 times, most recently
from
32e2513
to
0e6e506
Compare
cxxly
changed the title
cxxly
force-pushed
the
spmd_elementwise
branch
4 times, most recently
from
bf27068
to
8f37184
Compare
cxxly
force-pushed
the
spmd_elementwise
branch
4 times, most recently
from
fa0f150
to
9382027
Compare
* fused_linear_param_grad_add_pass * modify * skip cpu ci * add muti_presion modify * modify * modify * modify * modify
cxxly
force-pushed
the
spmd_elementwise
branch
from
9382027
to
1d8a01e
Compare
cxxly
changed the title
cxxly
changed the title
JZ-LIANG
reviewed
Oct 31, 2023
| @@ -30,6 +30,10 @@ SpmdInfo ElementwiseUnaryInferSpmdReverse(const DistMetaTensor& x, | |||
| SpmdInfo ElementwiseUnaryGradInferSpmd(const DistMetaTensor& x, | |||
| const DistMetaTensor& out_grad); | |||
|
|
|||
| SpmdInfo ElementwiseUnaryGradInferSpmd(const DistMetaTensor& x, | |||
There was a problem hiding this comment.
缺少单规则模块的单测，可以参考 test/auto_parallel/spmd_rules/test_elementwise_rule.py
| @@ -42,5 +46,11 @@ SpmdInfo ElementwiseBinaryGradInferSpmd(const DistMetaTensor& x, | |||
| const DistMetaTensor& out_grad, | |||
| int64_t axis = -1); | |||
|
|
|||
| SpmdInfo ElementwiseBinaryGradInferSpmd(const DistMetaTensor& x, | |||
pkuzyc
reviewed
Oct 31, 2023
| @@ -30,6 +30,10 @@ SpmdInfo ElementwiseUnaryInferSpmdReverse(const DistMetaTensor& x, | |||
| SpmdInfo ElementwiseUnaryGradInferSpmd(const DistMetaTensor& x, | |||
| const DistMetaTensor& out_grad); | |||
|
|
|||
| SpmdInfo ElementwiseUnaryGradInferSpmd(const DistMetaTensor& x, | |||
There was a problem hiding this comment.
重载是为了适配 api 么？是不是可以加些注释说明一下
There was a problem hiding this comment.
适配不同API签名，比如一元逐元素运算，部分算子反向只依赖输入，部分依赖输入输出，sin_grad(const Tensor& x, const Tensor& out_grad, Tensor* x_grad), silu_grad(const Tensor& x, const Tensor& out, const Tensor& out_grad, Tensor* x_grad)
JZ-LIANG
approved these changes
Oct 31, 2023
|
你的PR提交成功，感谢你对开源项目的贡献! |
zeroRains
pushed a commit
to zeroRains/Paddle
that referenced
this pull request
Nov 8, 2023
…er semi auto parallel (PaddlePaddle#58474) * add elementwise spmd rules for eager semi auto parallel * support bool dtype for s_to_r_reshared * increase test_semi_auto_parallel_basic timeout for pass ci * prompt rtol tolerance from 1e-5 to 1e-6 --------- Co-authored-by: xiaoguoguo626807 <100397923+xiaoguoguo626807@users.noreply.github.com>
danleifeng
pushed a commit
to danleifeng/Paddle
that referenced
this pull request
Nov 14, 2023
…er semi auto parallel (PaddlePaddle#58474) * add elementwise spmd rules for eager semi auto parallel * support bool dtype for s_to_r_reshared * increase test_semi_auto_parallel_basic timeout for pass ci * prompt rtol tolerance from 1e-5 to 1e-6 --------- Co-authored-by: xiaoguoguo626807 <100397923+xiaoguoguo626807@users.noreply.github.com>
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
New features
PR changes
Others
Description
Pcard-73145
Support bool dtype for sharding
Add yaml config and unittest for follow ops, which are used in LLaMa2 model.