Tag
JetSpec introduces parallel tree drafting for speculative decoding, achieving up to 9.64x end-to-end speedup on LLM inference while maintaining lossless accuracy, with throughput reaching ~1000 TPS on a single B200 GPU.
JetFlow is a speculative decoding framework that breaks the scaling ceiling by combining one-forward drafting efficiency with branch-wise causal conditioning, achieving up to 9.64x speedup on math benchmarks and outperforming prior methods on dense and MoE Qwen3 models.