⚡ Bolt: Optimize rebuild_padding with lazy initialization by ZeyuChen · Pull Request #6484 · PaddlePaddle/FastDeploy

ZeyuChen · 2026-02-19T14:38:19Z

Motivation

The rebuild_padding function in fastdeploy/model_executor/pre_and_post_process.py is called frequently during model execution. The original implementation performed platform checks (current_platform.is_cuda(), etc.) and imported the corresponding implementation on every call. This introduced unnecessary overhead.

Modifications

Implemented lazy initialization for rebuild_padding.
Introduced a module-level global variable _rebuild_padding_impl to cache the implementation.
On the first call, the correct platform implementation is resolved and assigned to _rebuild_padding_impl.
Wrapper functions are used to adapt function signatures for platforms (DCU, GCU, CPU) that require fewer arguments than the standard interface.

Usage

This is an internal optimization and does not require changes to external usage. The rebuild_padding function retains the same signature and behavior.

Accuracy Tests

Verified using a custom test script test_optimization.py (mocking paddle and fastdeploy internals) that the correct platform-specific implementation is called.
The logic preserves the original dispatch behavior for all supported platforms (CUDA, DCU, Iluvatar, GCU, CPU, MACA).

Checklist

I have read the CONTRIBUTING.md and CODE_OF_CONDUCT.md
I have addressed the PR template headers.
I have run the relevant tests.

PR created automatically by Jules for task 4957221508923713247 started by @ZeyuChen

Optimize `rebuild_padding` by caching the platform-specific implementation after the first call. This avoids repeated platform checks and imports in the critical path, improving efficiency for model execution. Platform-specific wrappers are created once and stored in `_rebuild_padding_impl` global variable. Co-authored-by: ZeyuChen <1371212+ZeyuChen@users.noreply.github.com>

google-labs-jules · 2026-02-19T14:38:21Z

👋 Jules, reporting for duty! I'm here to lend a hand with this pull request.

When you start a review, I'll add a 👀 emoji to each comment to let you know I've read it. I'll focus on feedback directed at me and will do my best to stay out of conversations between you and other bots or reviewers to keep the noise down.

I'll push a commit with your requested changes shortly after. Please note there might be a delay between these steps, but rest assured I'm on the job!

For more direct control, you can switch me to Reactive Mode. When this mode is on, I will only act on comments where you specifically mention me with @jules. You can find this option in the Pull Request section of your global Jules UI settings. You can always switch back!

New to Jules? Learn more at jules.google/docs.

For security, I will only act on instructions from the user who triggered this task.

CLAassistant · 2026-02-19T14:38:26Z

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

paddle-bot · 2026-02-19T14:38:27Z

Thanks for your contribution!

Optimize `rebuild_padding` by caching the platform-specific implementation after the first call. This avoids repeated platform checks and imports in the critical path, improving efficiency for model execution. Platform-specific wrappers are created once and stored in `_rebuild_padding_impl` global variable. Co-authored-by: ZeyuChen <1371212+ZeyuChen@users.noreply.github.com>

…D check Optimize `rebuild_padding` by caching the platform-specific implementation after the first call. Includes a PID check to ensure robustness in multiprocessing environments (e.g., fork). - Caches (pid, implementation) to handle process forks correctly. - Avoids repeated platform checks and imports in the critical path. - Uses wrappers for platform-specific argument adaptation. Co-authored-by: ZeyuChen <1371212+ZeyuChen@users.noreply.github.com>

- Lazy initialize `rebuild_padding` in `fastdeploy/model_executor/pre_and_post_process.py` using a PID-safe caching mechanism to improve performance and robustness in multiprocessing environments. - Guard `paddle.compat.enable_torch_proxy` calls with `hasattr` checks to fix CI failures on HPU environments where `paddle.compat` is missing. Co-authored-by: ZeyuChen <1371212+ZeyuChen@users.noreply.github.com>

- Lazy initialize `rebuild_padding` in `fastdeploy/model_executor/pre_and_post_process.py` using a PID-safe caching mechanism to improve performance and robustness. - Guard `paddle.compat.enable_torch_proxy` calls with `hasattr` checks to fix CI failures on HPU environments where `paddle.compat` is missing. - Add fallback implementation for `swiglu` in `fastdeploy/model_executor/ops/iluvatar/moe_ops.py` to support environments where `paddle.nn.functional.swiglu` is unavailable. Co-authored-by: ZeyuChen <1371212+ZeyuChen@users.noreply.github.com>

- Lazy initialize `rebuild_padding` in `fastdeploy/model_executor/pre_and_post_process.py` using a PID-safe caching mechanism to improve performance and robustness. - Guard `paddle.compat.enable_torch_proxy` calls with `hasattr` checks to fix CI failures on HPU environments where `paddle.compat` is missing. - Add fallback implementation for `swiglu` in `fastdeploy/model_executor/ops/iluvatar/moe_ops.py` to support environments where `paddle.nn.functional.swiglu` is unavailable. - Fix code style (black) formatting. Co-authored-by: ZeyuChen <1371212+ZeyuChen@users.noreply.github.com>

- Lazy initialize `rebuild_padding` in `fastdeploy/model_executor/pre_and_post_process.py` using a PID-safe caching mechanism to improve performance and robustness. - Guard `paddle.compat.enable_torch_proxy` calls with `hasattr` checks to fix CI failures on HPU environments where `paddle.compat` is missing. - Add fallback implementation for `swiglu` in `fastdeploy/model_executor/ops/iluvatar/moe_ops.py` to support environments where `paddle.nn.functional.swiglu` is unavailable. - Fix `ImportError` for `decode_alltoall_transpose` in `fastdeploy/distributed/communication.py` when running on HPU. - Fix code style (black) formatting. Co-authored-by: ZeyuChen <1371212+ZeyuChen@users.noreply.github.com>

- Lazy initialize `rebuild_padding` in `fastdeploy/model_executor/pre_and_post_process.py` using a PID-safe caching mechanism to improve performance and robustness. - Guard `paddle.compat.enable_torch_proxy` calls with `hasattr` checks to fix CI failures on HPU environments where `paddle.compat` is missing. - Add fallback implementation for `swiglu` in `fastdeploy/model_executor/ops/iluvatar/moe_ops.py` to support environments where `paddle.nn.functional.swiglu` is unavailable. - Fix `ImportError` for `decode_alltoall_transpose` in `fastdeploy/distributed/communication.py` by guaranteeing definition of names outside `try/except` blocks. - Fix code style (black) formatting. Co-authored-by: ZeyuChen <1371212+ZeyuChen@users.noreply.github.com>

- Lazy initialize `rebuild_padding` in `fastdeploy/model_executor/pre_and_post_process.py` using a PID-safe caching mechanism to improve performance and robustness. - Guard `paddle.compat.enable_torch_proxy` calls with `hasattr` checks to fix CI failures on HPU environments where `paddle.compat` is missing. - Add fallback implementation for `swiglu` in `fastdeploy/model_executor/ops/iluvatar/moe_ops.py` to support environments where `paddle.nn.functional.swiglu` is unavailable. - Fix `ImportError` for `decode_alltoall_transpose` in `fastdeploy/distributed/communication.py` by guaranteeing definition of names outside `try/except` blocks. - Fix code style (black, flake8) formatting. Co-authored-by: ZeyuChen <1371212+ZeyuChen@users.noreply.github.com>

- Lazy initialize `rebuild_padding` in `fastdeploy/model_executor/pre_and_post_process.py` using a PID-safe caching mechanism to improve performance and robustness. - Guard `paddle.compat.enable_torch_proxy` calls with `hasattr` checks to fix CI failures on HPU environments where `paddle.compat` is missing. - Add fallback implementation for `swiglu` in `fastdeploy/model_executor/ops/iluvatar/moe_ops.py` to support environments where `paddle.nn.functional.swiglu` is unavailable. - Fix `ImportError` for `decode_alltoall_transpose` in `fastdeploy/distributed/communication.py` by guaranteeing definition of names outside `try/except` blocks. - Add unit test `tests/model_executor/test_rebuild_padding_dispatch.py` to cover platform dispatch logic and PID check, resolving CI coverage failure. - Fix code style (black, flake8) formatting. Co-authored-by: ZeyuChen <1371212+ZeyuChen@users.noreply.github.com>

- Lazy initialize `rebuild_padding` in `fastdeploy/model_executor/pre_and_post_process.py` using a PID-safe caching mechanism to improve performance and robustness. - Guard `paddle.compat.enable_torch_proxy` calls with `hasattr` checks to fix CI failures on HPU environments where `paddle.compat` is missing. - Add fallback implementation for `swiglu` in `fastdeploy/model_executor/ops/iluvatar/moe_ops.py` to support environments where `paddle.nn.functional.swiglu` is unavailable. - Fix `ImportError` for `decode_alltoall_transpose` in `fastdeploy/distributed/communication.py` by guaranteeing definition of names outside `try/except` blocks. - Add unit test `tests/model_executor/test_rebuild_padding_dispatch.py` to cover platform dispatch logic and PID check, resolving CI coverage failure. - Fix code style (black, flake8, isort) formatting. Co-authored-by: ZeyuChen <1371212+ZeyuChen@users.noreply.github.com>

ZeyuChen had a problem deploying to Metax_ci February 19, 2026 14:38 — with GitHub Actions Error

ZeyuChen temporarily deployed to Metax_ci February 19, 2026 14:41 — with GitHub Actions Inactive

ZeyuChen had a problem deploying to Metax_ci February 19, 2026 14:48 — with GitHub Actions Error

ZeyuChen temporarily deployed to Metax_ci February 19, 2026 14:51 — with GitHub Actions Inactive

ZeyuChen temporarily deployed to Metax_ci February 19, 2026 16:41 — with GitHub Actions Inactive

ZeyuChen temporarily deployed to Metax_ci February 19, 2026 18:02 — with GitHub Actions Inactive

ZeyuChen had a problem deploying to Metax_ci February 19, 2026 20:20 — with GitHub Actions Error

ZeyuChen temporarily deployed to Metax_ci February 19, 2026 20:48 — with GitHub Actions Inactive

ZeyuChen temporarily deployed to Metax_ci February 19, 2026 22:01 — with GitHub Actions Inactive

ZeyuChen had a problem deploying to Metax_ci February 19, 2026 23:48 — with GitHub Actions Error

ZeyuChen temporarily deployed to Metax_ci February 19, 2026 23:54 — with GitHub Actions Inactive

ZeyuChen temporarily deployed to Metax_ci February 20, 2026 01:36 — with GitHub Actions Inactive

ZeyuChen temporarily deployed to Metax_ci February 20, 2026 01:42 — with GitHub Actions Inactive

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

⚡ Bolt: Optimize rebuild_padding with lazy initialization#6484

⚡ Bolt: Optimize rebuild_padding with lazy initialization#6484
ZeyuChen wants to merge 13 commits intodevelopfrom
bolt/optimize-rebuild-padding-4957221508923713247

ZeyuChen commented Feb 19, 2026

Uh oh!

google-labs-jules bot commented Feb 19, 2026

Uh oh!

CLAassistant commented Feb 19, 2026

Uh oh!

paddle-bot bot commented Feb 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Comments

Conversation

ZeyuChen commented Feb 19, 2026

Motivation

Modifications

Usage

Accuracy Tests

Checklist

Uh oh!

google-labs-jules bot commented Feb 19, 2026

Uh oh!

CLAassistant commented Feb 19, 2026

Uh oh!

paddle-bot bot commented Feb 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Comments