C500 FlashMLA支持

在第一时间我们支持FlashMLA，并开放代码至：
代码仓库地址：https://github.com/MetaX-MACA/FlashMLA
欢迎给我们提Iusse交流及贡献：https://github.com/MetaX-MACA/FlashMLA/issues


FlashMLA on MXMACA

We provide the implementation of FlashMLA from FlashAttention-2(version 2.6.3), based on MACA toolkit and C500 chips.

FlashAttention-2 currently supports:

• Datatype fp16 and bf16.
• Multi-Token Parallelism = 1
• Paged kvcache with block size equal to 2^n (n >= 0)
  

How to run on MXMACA DeviceInstallation

Requirements:

• MXMACA GPUs.
• MACA development toolkit.
• mcTlass source code.
• mcPytorch2.1 and mcTriton2.1 from maca toolkit wheel package and above.
  

To install:

• Make sure that maca pyTorch2.1 is installed.
• Download mctlass source code from: https://sw-download.metax-tech.com/