vLLM reports Reflection's DFlash2 speeds up Beam decoding 1.72–3.03×

1. 11:26 AM
   vLLM@vllm_projectOfficial

   DFlash2 speeds Beam's decode path 1.72–3.03× in Reflection tests

   AIReflection reports that DFlash2 speeds up Beam's decode path by 1.72× to 3.03× across GSM8K, MATH500, HumanEval, MBPP and MT-Bench at concurrency 1, 16 and 32. On GSM8K the speedups are 3.03×, 2.51× and 2.75× at those concurrency levels, measured in a disaggregated prefill/decode setup with prefill time excluded. Reflection also reports better throughput with K=7, although the drafter was trained with block size 16.

   
2. 9:23 AM
   vLLM@vllm_projectOfficial

   DFlash2 speeds up Beam decoding 1.72–3.03× in Reflection tests

   AIReflection reports that its DFlash2 drafter speeds up Beam's decode path by 1.72–3.03× across GSM8K, MATH500, HumanEval, MBPP and MT-Bench at concurrency 1, 16 and 32. On GSM8K the speedups reach 3.03×, 2.51× and 2.75× at those concurrency levels, measured in a disaggregated prefill/decode setup with prefill time excluded. Reflection also reports better throughput with K=7, although the drafter was trained with block size 16.

   

The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.