[Feature] support W8A8(FP8) and KV Cache FP8 for DeepSeek V2

[zhyncs]

Checklist

• 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
• 2. Please use English, otherwise it will be closed.

Motivation

As titled. Make DeepSeek V2 MLA Faster!

Related resources

No response

[fengyang95]

Is there a specific timeline for this?

[zhyncs]

Is there a specific timeline for this?

bmm fp8 has been implemented with flashinfer-ai/flashinfer#469
fp8 e5m2 kv cache has been implemented with #1204

Currently, there is no adaptation for DeepSeek V2 as we are focusing on other higher priority tasks. Expected to be completed within these few days.

[zhyncs]

done