Chi-Chih Chang comments

Repositories
Issues
Comments

Results 2 comments of


                                            Chi-Chih Chang

[Feature Request] Flexible MLA Kernel Supporting More `d`, `dv`, and `n_heads` Values

Hi @chengyupku, I have tried to work on this and the reference program of @JT-Ushio's requests is attached below. The major difference between MHA2MLA and MLA is that the RoPE...

[Feature Request] Flexible MLA Kernel Supporting More `d`, `dv`, and `n_heads` Values

Hi @chengyupku, thank you! Yesterday I started modifying [example_mla_decode.py](https://github.com/tile-ai/tilelang/blob/main/examples/deepseek_mla/example_mla_decode.py#L133) with a minimal change in mind—specifically, replacing the GEMM operation between `Q_pe_shared` and `K_pe_shared` with a Batched GEMV instead. Here's the...