【pir】update If_grad and while_grad op' stop gradient by yield_op #70545
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
xiaoguoguo626807
requested review from
changeyoung98 and
phlrain
as code owners
|
你的PR提交成功，感谢你对开源项目的贡献! |
This was referenced Jan 2, 2025
github-merge-queue bot
pushed a commit
to deepmodeling/deepmd-kit
that referenced
this pull request
Jan 13, 2025
1. support LAMMPS inference with Paddle backend on cuda and dcu device 2. Fix `deepmd/pd/utils/serialization.py` and paddle inference files can be deserialized using `dp convert-backend deeppot_sea.yaml deeppot_sea.json` related PR: 1. PaddlePaddle/Paddle#70545 <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit Based on the comprehensive changes, here are the updated release notes: - **New Features** - Added support for PaddlePaddle deep learning framework. - Introduced new Paddle-based deep potential model computations. - Expanded backend support for model inference with Paddle. - **Configuration** - New CMake option `ENABLE_PADDLE` to toggle Paddle support. - Added configuration parameters for Paddle version and inference directory. - **Testing** - Comprehensive test suites added for Paddle backend. - Enhanced LAMMPS integration tests with Paddle support. - **Documentation** - Updated version header and configuration files to reflect Paddle integration. - **Performance** - Added JIT compilation for Paddle model methods. - Optimized model serialization and deserialization. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: HydrogenSulfate <[email protected]> Signed-off-by: Jinzhe Zeng <[email protected]> Co-authored-by: Jinzhe Zeng <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
github-merge-queue bot
pushed a commit
to deepmodeling/deepmd-kit
that referenced
this pull request
Feb 8, 2025
We verified paddle CINN compiler in DPA-2 example(single A100-SXM (40G), cada11.8, Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz x 160). To enable CINN compiler in training, add one flag: `CINN=1` before training command, e.g. `CINN=1 dp --pd train input_torch_medium.json`. ### Curves: #### dpa2  ##### Performance We tested with torch==2.6.0.dev20241219+cu118 | Pytorch(eager) | Pytorch(compile) | Paddle(eager) | Paddle(CINN compile) | | -- | -- | -- | -- | | 0.1078s/step | compile do not support double-backward | 0.1130s/step | 0.0638s/step | #### se_atten  ##### Performance We tested with torch==2.6.0.dev20241219+cu118 | Pytorch(eager) | Pytorch(compile) | Paddle(eager) | Paddle(CINN compile) | | -- | -- | -- | -- | | 0.0342s/step | compile do not support double-backward | 0.0444s/step | 0.0244s/step | ### Accuracy details: #### dpa2 - Pytorch:  - Paddle(eager mode):  - Paddle(CINN compiler)  #### se_atten - Pytorch  - Paddle(eager_mode)  - Paddle(CINN compliler)  ### TODO: - [x] Fix high order backward of control flow in Paddle(PaddlePaddle/Paddle#70545) - [ ] Fix shape64 issue(PaddlePaddle/Paddle#70947) <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit - **New Features** - Enhanced training performance through optimization techniques that leverage compilation and profiling for better execution monitoring. - Introduced a new function for converting various input types to boolean values, improving environment variable handling. - **Refactor** - Improved precision management for training parameters and updated environment configuration handling to ensure robust optimization support. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@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 Category
Execute Infrastructure
PR Types
Bug fixes
Description
pcard-67164
对于控制流高阶微分，其一阶反向是先创建空op, 再单独向block 中添加子op, 添加之后需要根据里面op 输出的stop_gradient 进行更新