LiLiCorr correlates parallel drafts for speculative decoding, ~20% throughput gain
eliebakouch · x · 2026-10-15
eliebakouch shares the LiLiCorr paper (arXiv:2608.20530) behind Nemotron 3.5 Bolt's parallel drafter. It jointly scores top-K candidates across an entire draft block in one lightweight network pass via in/out vector cosine alignment, avoiding DFlash's jointly incoherent marginals and DSpark's token-by-token correction. Results: 10% better acceptance rate at equal training in short-context and 20% better long-context throughput vs DSpark, speeding up 24GB local inference of the 74B-A9B model.
More from Infra
- SGLang ships Day-0 support for FLUX 3 Image Klein: 10 reference images, BF16/FP8/NVFP4 variants — ying11231 · 2026-10-16
- NVIDIA's Nemotron 3.5 Super VL hits CoreWeave: 12B-active multimodal MoE at $0.25/M input tokens — wandb · 2026-10-16
- Chip equipment beats bull case: 2026 WFE seen at $153B, $278B by 2030 — BenBajarin · 2026-10-16
- Steam now launches in 5.23s, beating Epic's 5.47s — and vLLM gets roasted for slow startup — AlpinDale · 2026-10-16
- Shadeform to Announce Six New Features on Oct 19: Lower GPU Costs, More Suppliers — _jaydeepkarale · 2026-10-16
- Fastino launches Training API for SFT/RL, claiming 25-85% cost cut vs Tinker — kalyan_kpl · 2026-10-16