Is there context parallelism method using exact algorithm for MLA prefill in pytorch backend?

[ariesjzj]

There are four types of CP:

• ULYSSES: not supported for MLA
• STAR: approximation algorithm, not an exact one
• RING: only available in TensorRT backend
• HELIX: suitable for decode instead of prefill

Is there any CP method for prefill in pytorch backend, for models such as DeepSeek V3 & DeepSeek V3.2?
If no, is there any plan for it? Thanks.