[Deterministic] Move paddle version batch invariant pkg to Fastdeploy #4763
+933
−0
Conversation
This file contains hidden or 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
|
Thanks for your contribution! |
littledgg
changed the title
22 tasks
Collaborator
|
please format you code |
Contributor
Author
done |
Contributor
There was a problem hiding this comment.
Pull Request Overview
This PR introduces batch-invariant implementations of key PaddlePaddle operations (mm, addmm, log_softmax, mean) using Triton kernels to achieve deterministic inference results regardless of batch size. The implementation is adapted from the batch_invariant_ops library and integrated into FastDeploy.
- Adds custom Triton kernel implementations for deterministic matrix operations and reduction operations
- Provides a context manager to toggle between standard and batch-invariant modes
- Includes comprehensive test files demonstrating batch invariance for each operation
Reviewed Changes
Copilot reviewed 6 out of 6 changed files in this pull request and generated 18 comments.
Show a summary per file
| File | Description |
|---|---|
| fastdeploy/model_executor/layers/batch_invariant_ops/batch_invariant_ops.py | Core implementation with Triton kernels for batch-invariant operations and mode switching functionality |
| fastdeploy/model_executor/layers/batch_invariant_ops/init.py | Module initialization exporting public API |
| tests/batch_invariant/test_batch_invariance_op_mm.py | Test suite for matrix multiplication batch invariance |
| tests/batch_invariant/test_batch_invariance_op_mean.py | Test suite for mean operation batch invariance |
| tests/batch_invariant/test_batch_invariance_op_logsoftmax.py | Test suite for log_softmax operation batch invariance |
| tests/batch_invariant/test_batch_invariance_op_addmm.py | Test suite for addmm operation batch invariance |
fastdeploy/model_executor/layers/batch_invariant_ops/batch_invariant_ops.py
Outdated
Show resolved
Hide resolved
fastdeploy/model_executor/layers/batch_invariant_ops/batch_invariant_ops.py
Show resolved
Hide resolved
fastdeploy/model_executor/layers/batch_invariant_ops/batch_invariant_ops.py
Show resolved
Hide resolved
fastdeploy/model_executor/layers/batch_invariant_ops/batch_invariant_ops.py
Show resolved
Hide resolved
fastdeploy/model_executor/layers/batch_invariant_ops/batch_invariant_ops.py
Show resolved
Hide resolved
…ariant_ops.py 存在于原版代码注释中的版本控制遗留的内容，确实应该去除 Co-authored-by: Copilot <[email protected]>
Co-authored-by: Copilot <[email protected]>
Co-authored-by: Copilot <[email protected]>
…ariant_ops.py Co-authored-by: Copilot <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
Motivation
Achieving batch invariance in the PaddlePaddle framework.
Batch invariance：https://thinkingmachines.ai/blog/defeating-nondeterminism-in-llm-inference/
想要跑通需要安装如下内容，paddle必须是比较新的(建议用最新的)
如果能看见Batch-Invariant Mode下均为0就代表正确
目前只有log_softmax算子尽管精心构造了输入数据，但是在原版实现似乎就已经具备批处理不变性了。
TODO：严格对齐API目前(mm和log_softmax还存在问题)，可以考虑把test case整合进一个文件,文件中列出的若干TODO
Modifications
Usage or Command
Accuracy Tests
Checklist
[FDConfig],[APIServer],[Engine],[Scheduler],[PD Disaggregation],[Executor],[Graph Optimization],[Speculative Decoding],[RL],[Models],[Quantization],[Loader],[OP],[KVCache],[DataProcessor],[BugFix],[Docs],[CI],[Optimization],[Feature],[Benchmark],[Others],[XPU],[HPU],[GCU],[DCU],[Iluvatar],[Metax]]pre-commitbefore commit.releasebranch, make sure the PR has been submitted to thedevelopbranch, then cherry-pick it to thereleasebranch with the[Cherry-Pick]PR tag.