Pinned post
Introducing JetSpec++: we find causal parallel drafting paired with hidden states predictor and anchor token decoder can effectively draft the next future block before the current one is verified, and spending the freed time on self refinement.
This allows drafting to be pipelined with verification, hiding almost all of its drafting cost.
JetSpec++ reaches up to 1.54x end-to-end speedup over DSpark and 7.45x over autoregressive decoding while keeping lossless, at 1,361 TPS on Qwen3-8B and 367 TPS on Qwen3.8-27B in BF16.
With NVFP4 and kernel optimizations, JetSpec++ further reaches over 4,000 TPS on Qwen3-8B and over 900 TPS on Qwen3.8-27B, on two GB200 GPUs. ⚡️
Project page: jetspec-project.github.io/jetspecpp-web/
This allows drafting to be pipelined with verification, hiding almost all of its drafting cost.
JetSpec++ reaches up to 1.54x end-to-end speedup over DSpark and 7.45x over autoregressive decoding while keeping lossless, at 1,361 TPS on Qwen3-8B and 367 TPS on Qwen3.8-27B in BF16.
With NVFP4 and kernel optimizations, JetSpec++ further reaches over 4,000 TPS on Qwen3-8B and over 900 TPS on Qwen3.8-27B, on two GB200 GPUs. ⚡️
Project page: jetspec-project.github.io/jetspecpp-web/
1:26
