Weekly Meeting 2026‐01‐23

Issues

To be discussed:

Development and testing tool chains:

• [Enhancement] Replace CSV-Based Benchmark Configuration with Pytest-Native Approach

• [Enhancement] Adopt a Modern Python Development Toolchain

• [Enhancement] Adopt pytest for Unit Testing

• [Discussion] Developing Kernels Based on TileLang Eager Mode

• [Potential] The benchmark check's default constraint is too relaxed.

Technical problems:

• [Feature Request] Kernel workspace interface for higher-level APIs

• [Feature Request] Kernel cache

Assigned

To be solved:

MHA @YuchongLi

• [Feature Request] Multi-Head-Attention - Benchmark

CI @YuchongLi

• [Enhancement] Add Docker Images in CI
• [CI][Enhancement] Miss the tileOps install and build process
• [CI][Enhancement] Refactor CI functionality to enhance CI stability and expand test coverage

NSA @JienengYu

• NSA_fwd
• nsa_topk

MLA decode with kvcache @ZhejinXu

• [Feature Request] MultiHeadLatentAttentionDecodeWithKVCache Kernels
• [Feature Request] MultiHeadLatentAttentionDecodeWithKVCache Function
• [Feature Request] MultiHeadLatentAttentionDecodeWithKVCache Function
• [Feature Request] MultiHeadLatentAttentionDecodeWithKVCache Layer

manifold-constrained hyper-connection @AngGao

• [New Op] manifold Hyper Connection (mHC)

DeepSeek Sparse Attention Decode @YuxianDu

• [feature Request] Implement Indexer Kernel for DSA Attention in TileOps
• [Feature] [New Op] implement topk op for DSA

Solved this week:

Decouple Submodules Dependency @YuchongLi

• The benchmarks/init.py unconditionally imports all submodules
• [Enhancement] Move imports of operator baseline dependency libraries inside the baseline function

PRs:

Merged:

• [Refactor] Refactor Gemm code format @YuchongLi
• [CI][Enhancement] Hide private environment variables in the ci.yml file @YuchongLi
• [CI][Bug] Fix Bug in CI conda @YuchongLi
• [Enhancement] Move imports of MHA/GQA benchmarks baseline dependency libraries inside the baseline function. @YuchongLi
• [CI][Enhancement] Delete flake8 from pre-commit hook. @YuchongLi
• [refactor][nsa] Decouple from third-party libraries @JienengYu

Actions Items

to-do

1. Create documentation for user-facing dependency installation
2. Convert Python tests to Pytest.
3. Distribute Docker images (one for internal development, one for external reproduction).

done

1. Decouple benchmark dependencies. [Enhancement] Move imports of MHA/GQA benchmarks baseline dependency libraries inside the baseline function. @YuchongLi