Set Block Decoding is a Language Model Inference Accelerator
This addresses inference efficiency for language model deployment, offering a flexible acceleration method without architectural changes.
The paper tackles the high computational and memory costs of inference in autoregressive language models by introducing Set Block Decoding (SBD), which accelerates generation by integrating next token and masked token prediction to sample multiple future tokens in parallel, resulting in a 3-5x reduction in forward passes while maintaining performance.
Autoregressive next token prediction language models offer powerful capabilities but face significant challenges in practical deployment due to the high computational and memory costs of inference, particularly during the decoding stage. We introduce Set Block Decoding (SBD), a simple and flexible paradigm that accelerates generation by integrating standard next token prediction (NTP) and masked token prediction (MATP) within a single architecture. SBD allows the model to sample multiple, not necessarily consecutive, future tokens in parallel, a key distinction from previous acceleration methods. This flexibility allows the use of advanced solvers from the discrete diffusion literature, offering significant speedups without sacrificing accuracy. SBD requires no architectural changes or extra training hyperparameters, maintains compatibility with exact KV-caching, and can be implemented by fine-tuning existing next token prediction models. By fine-tuning Llama-3.1 8B and Qwen-3 8B, we demonstrate that SBD enables a 3-5x reduction in the number of forward passes required for generation while achieving same performance as equivalent NTP training.