fep(sig-operator): add FlagGems-vllm high-performance fused operator library proposal#20
Open
huangyiqun wants to merge 5 commits into
Open
fep(sig-operator): add FlagGems-vllm high-performance fused operator library proposal#20huangyiqun wants to merge 5 commits into
huangyiqun wants to merge 5 commits into
Conversation
huangyiqun
changed the title
zckzck
reviewed
May 28, 2026
|
|
||
| This design allows the same operator API to be used across supported hardware backends as implementations become available. | ||
|
|
||
| ### Testing and Benchmarking |
There was a problem hiding this comment.
Test: Does FlagGems-vllm only support NVIDIA hardware, or does it work with other vendors? If compatible, please list the supported vendors.
Author
There was a problem hiding this comment.
The multi backend adaptation and verification are currently underway.
zckzck
reviewed
May 28, 2026
| | Dedicated package import | Run `python -c "import flaggems_vllm; import flaggems_vllm.ops"` after installation. | | ||
| | Fused operator API availability | Verify exported symbols from `flaggems_vllm.ops.__all__` include the migrated vLLM-facing fused operators. | | ||
| | Accuracy coverage | Run `pytest -q tests --collect-only` and targeted tests such as `pytest -q tests/test_moe_align_block_size.py --quick`. | | ||
| | DeepSeek V4 helper coverage | Run the DeepSeek V4 attention helper tests when the matching CUDA/vLLM reference environment is available. | |
There was a problem hiding this comment.
Specific test methods & test procedures ?
Author
There was a problem hiding this comment.
The specific testing methods have been added.
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
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.
FEP: FlagGems-vllm
Adds a FEP document for FlagGems-vllm, a high-performance fused operator library for vLLM inference workloads in the FlagOS ecosystem.
FlagGems-vllm provides Triton-based fused kernels and vLLM-facing operator implementations for performance-critical paths such as MoE routing, cache update, rotary embedding, FP8 quantization, sequence pack/unpack, and DeepSeek V4 attention helper kernels.
The FEP defines the repository scope, fused operator coverage, packaging approach, test plan, and migration process for keeping vLLM-related fused kernels in sync with FlagGems while exposing them through the standalone
flaggems_vllmpackage.Repository: https://github.com/flagos-ai/FlagGems-vllm