How are you doing sparse attention for rtx blackwell? sm120

### Reminder

- [x] I have read the above rules and searched the existing issues.

### System Info

this repo on HF Kebob/DeepSeek-V3.2-CPU-NUMA2-AMXINT4
claims to run this on RTX PRO 6000 sm120 and ktransformers
but flashmla the only sparse attention backend im aware of doesn't support sm120??
and sglang instructions for ds32 include installing flashmla...?

### Reproduction

```text
Put your message here.
```

### Others

_No response_