speculative decoding
speculative-decoding-be387a47·14 events·first seen 1mo agoAliases: speculative decoding, self-speculative decoding
Co-occurring entities
More like this (12)
Guides (1)
Recent events (14)
Speculative Decoding for 2x Faster Whisper Inference
Hugging Face demonstrates applying speculative decoding to OpenAI's Whisper speech recognition model, achieving approximately 2x inference speedup. The technique uses a smaller draft model to propose token sequences that the larger target model then verifies, reducing the number of full forward passes required. This post covers implementation details using the Hugging Face Transformers library and benchmarks the approach across different hardware configurations.
Faster Text Generation with Self-Speculative Decoding via LayerSkip
This Hugging Face blog post covers LayerSkip, a self-speculative decoding technique that accelerates text generation by using early exit from transformer layers to draft tokens, then verifying them with the full model. Unlike standard speculative decoding, LayerSkip requires no separate draft model, reducing memory overhead while still achieving inference speedups. The post likely covers integration with the Hugging Face ecosystem and practical performance benchmarks.
Assisted Generation: a new direction toward low-latency text generation
Hugging Face introduces assisted generation (speculative decoding) as a practical technique for reducing LLM inference latency. The approach uses a smaller draft model to propose token candidates that a larger model then verifies in parallel, enabling multiple tokens to be accepted per forward pass. The blog post explains the mechanism and demonstrates integration into the Hugging Face Transformers library.
Graft: Hybrid Tree Construction for Speculative Decoding via Prune-Then-Retrieve
Graft is a training-free framework that improves speculative decoding by coupling dynamic-depth pruning with retrieval-based token compensation. Pruning reduces VRAM and compute overhead while freeing budget for retrieval, which fills topological gaps in the draft tree with near-zero additional cost. On short-context benchmarks, Graft achieves up to 5.41× speedup and improves average speedup over EAGLE-3 by up to 21.8% on Qwen3-235B. The method is evaluated across short- and long-context settings and extended to block-drafting paradigms.
SimSD: Speculative Decoding Adapted for Diffusion Language Models
SimSD introduces a training-free speculative decoding algorithm for diffusion large language models (dLLMs), which previously could not use standard token-level speculative decoding due to their bidirectional attention and masked language modeling formulation. The method uses a plug-and-play masking strategy that introduces reference tokens from a draft model and a custom attention mask, enabling valid logit computation for drafted tokens in a single forward pass. Evaluated on SDAR-family dLLMs across four benchmarks, SimSD achieves up to 7.46x decoding throughput improvement while maintaining or improving generation quality. The approach is compatible with other acceleration techniques such as KV cache and blockwise decoding.
VIA-SD: Multi-tier speculative decoding via intra-model routing cuts rejection rates and boosts inference speed
VIA-SD introduces a three-tier verification framework for speculative decoding that routes draft tokens to a lightweight 'slim verifier' submodel for medium-confidence cases, reserving full-model verification only for uncertain tokens. Across four tasks and multiple model families, the method reduces rejection rates by 0.10–0.22 and achieves 10–20% speedups over strong speculative decoding baselines, with 2.5–3x acceleration over standard decoding. The approach is compatible with existing speculative decoding frameworks without retraining. The work proposes multi-tier speculative decoding as a general paradigm for scalable LLM inference.
Faster Assisted Generation Support for Intel Gaudi
Hugging Face has published a blog post detailing assisted generation (speculative decoding) support optimized for Intel Gaudi accelerators. The post covers implementation details and performance improvements achieved by running assisted/speculative decoding on Gaudi hardware. This represents an infrastructure and inference optimization development relevant to non-NVIDIA AI accelerator deployment.
Universal Assisted Generation: Faster Decoding with Any Assistant Model
Hugging Face introduces Universal Assisted Generation (UAG), a technique that extends speculative decoding to work with any assistant model regardless of tokenizer or vocabulary differences. The approach enables using smaller, mismatched draft models to accelerate inference of larger target models, removing the previous constraint that both models share the same tokenizer. This broadens the practical applicability of speculative decoding across the open-weights ecosystem.
Faster Assisted Generation with Dynamic Speculation
Hugging Face introduces dynamic speculation lookahead for assisted (speculative) decoding, a technique that adaptively adjusts the number of candidate tokens generated by a draft model before verification by the main model. This approach aims to improve throughput and reduce latency compared to fixed-lookahead speculative decoding by tuning the speculation depth at runtime. The blog post describes the method and its integration into the Hugging Face Transformers library.
Accelerating Qwen3-8B Agent on Intel Core Ultra with Depth-Pruned Draft Models
Hugging Face and Intel demonstrate speculative decoding acceleration for the Qwen3-8B model on Intel Core Ultra client hardware using depth-pruned draft models. The approach applies structured pruning to create a smaller draft model that enables speculative decoding, targeting on-device agent workloads. This work addresses inference efficiency for mid-size open-weight models on consumer-grade x86 silicon.
Powerful ASR + Diarization + Speculative Decoding with Hugging Face Inference Endpoints
Hugging Face published a blog post describing a pipeline that combines automatic speech recognition (ASR), speaker diarization, and speculative decoding on their Inference Endpoints platform. The post demonstrates how these three techniques can be integrated to produce faster, speaker-attributed transcriptions. Speculative decoding is highlighted as a key inference optimization that reduces latency for ASR workloads.
Accelerate StarCoder with Optimum Intel on Xeon: Q8/Q4 and Speculative Decoding
Hugging Face and Intel demonstrate quantization (INT8/INT4) and speculative decoding techniques applied to StarCoder on Intel Xeon CPUs using the Optimum Intel library. The post covers practical inference acceleration workflows targeting CPU deployment of code generation models. This represents a concrete inference-economics use case for open-weight code models on commodity server hardware.
Pair-In, Pair-Out (PIPO): Unified Latent Compression and Multi-Token Prediction for Efficient LLM Inference
PIPO is a new inference efficiency framework that unifies input-side latent compression with output-side multi-token prediction (MTP) by treating them as mirror operations: a compressor folds two input tokens into one latent, while an MTP head unfolds one hidden state into an additional output token. To avoid the expensive verifier pass typically required by speculative decoding, PIPO trains a lightweight confidence head using On-Policy Distillation (OPD), which naturally aligns with rejection-sampling criteria. Experiments on Qwen3.5-4B and 9B backbones across AIME 2025, GPQA-Diamond, LiveCodeBench v6, and LongBench v2 show up to 2.64× first-token-latency speedup and +7.15 pass@4 improvement over regular decoding.
Bebop: MTP with rejection sampling and TV loss achieves 1.8x RL training speedup
Researchers introduce Bebop, a framework for integrating Multi-Token Prediction (MTP) into large-scale RL training pipelines for LLMs. The work identifies that MTP acceptance rates degrade during RL due to entropy fluctuations, and proposes probabilistic rejection sampling plus a novel end-to-end Total Variation (TV) loss that directly optimizes multi-step acceptance rates, achieving up to 95% acceptance rates and 25% extra inference throughput gains. Applied to Qwen3.5, Qwen3.6, and Qwen3.7 models, the method yields up to 1.8x end-to-end acceleration in async RL training. The approach eliminates the need for costly online MTP updating by using pre-RL MTP training with the proposed objectives.