Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
26 lines (21 loc) · 1.8 KB

CONTRIBUTING.md

File metadata and controls

26 lines (21 loc) · 1.8 KB
Raw

Welcome to sgl-kernel-xpu! We are building high performance kernels for intel GPUs. Appreciate your interest in contributing.This documentation provides a general guidelines for how to open a Pull Request.

We encourage you to open Pull Request from the following 4 perspectives:

  • Feature Enabling: Enable a new feature for XPU device. For example, enable fused_add_rmsnorm for XPU.
  • Performance Optimization: Improve the performance on XPU for critical operators, data types or quantization recipes. For example, support FP8 Flash Attention.
  • Module Refactoring: Enhance the maintainability and readability of the code. For example, refactor the scratchpad memory allocation in kernels.
  • Bug Fixing: Enhance test case coverage and code robustness.

Guidelines for Pull Requests:

  • Make sure update test cases when enabling a new feature and comprehesively cover all possible combinaitons in the test suite.
  • Make sure update benchmarks cases when enabling a new feature or optimizing for new scenario especially for critical operators.
  • Provide benchmark data to prove affectiveness of performance optimizations.
  • Use main branch for device-agnostic features and optimizations.
  • DO NOT expose hardware or software features for Intel(R) Protect IP on main branch.

Review Process for Merging Pull Requests:

  • Mark Pull Requests as draft and ask review from maintainers (mingfeima, airMeng, chunyuan-w, sunjiweiswift)
  • At least one approval is needed before landing for simple Pull Requests
  • At least two approvals are needed before landing for complexed Pull Requests.
  • Mark Pull Requests as ready for review once comments are addressed.
  • Pass all CI tests.

Extra Requirements:

  • For public Pull Requests on sglang repo, conduct internal review first before requesting reviews from community.