DiffuSpec Boosts Speculative Decoding Using Diffusion Models

DiffuSpec swaps the usual draft model for a pretrained diffusion language model, a training‑free method that can cut inference time by up to three times, per benchmark results. Read more: getnews.me/diffuspec-boosts-specula... #diffuspec #speculativdecoding

SelfJudge Boosts Speculative Decoding for Faster LLM Inference

SelfJudge trains verification judges from the model’s own outputs, removing the need for human‑annotated data. Benchmarks show faster inference with higher accuracy. Read more: getnews.me/selfjudge-boosts-specula... #selfjudge #speculativdecoding

Side-Channel Risks in Speculative Decoding for Large Language Models

Researchers found a side‑channel in speculative decoding that can fingerprint queries with up to 100% accuracy on REST and leak data at over 25 tokens per second. Read more: getnews.me/side-channel-risks-in-sp... #speculativdecoding #sidechannel

FastGRPO: Speeding Policy Optimization with Concurrency‑Aware Decoding

FastGRPO speeds up GRPO training by 2.35×‑2.72× using concurrency‑aware speculative decoding and online draft learning, keeping reasoning quality stable. Read more: getnews.me/fastgrpo-speeding-policy... #fastgrpo #speculativdecoding #rl

Training‑Free Speculative Decoding Enhances LLaMA 3 Speed and Accuracy

Training‑free speculative decoding lifts LLaMA 3 scores by 3.3 points and speeds generation 2.23×, with draft token acceptance up to 2.39 tokens. Read more: getnews.me/training-free-speculativ... #llama3 #speculativdecoding