Any plan to support suffix decoding for async scheduling?

[charlotte12l]

Currently, async scheduling is only supported with EAGLE/MTP kind of speculative decoding

https://github.com/vllm-project/vllm/blob/c4e744dbd41f23ee7fd554a86cc1bf516082552d/vllm/config/vllm.py#L574-L576

[sfc-gh-yewang]

Thanks for trying ArcticInference! Yes, we'll support in the v0.14.1 rebase https://github.com/snowflakedb/ArcticInference/tree/wangye/v14

[HF-001]

Thanks for trying ArcticInference! Yes, we'll support in the v0.14.1 rebase https://github.com/snowflakedb/ArcticInference/tree/wangye/v14

@sfc-gh-yewang hi, I tested the performance of suffix + async-scheduling on Sonnet data and found that compared to suffix + no-async-scheduling, the performance decreased. May I ask if this result is consistent with your testing?

[crystalww]

Any updates?