SlimSpec replaces the standard LM-head in draft models with a low-rank version to deliver 4-5x faster speculative decoding while preserving full vocabulary and competitive acceptance rates.
hub
Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads
23 Pith papers cite this work. Polarity classification is still indexing.
23
Pith papers citing it
abstract
Large Language Models (LLMs) employ auto-regressive decoding that requires sequential computation, with each step reliant on the previous one's output. This creates a bottleneck as each step necessitates moving the full model parameters from High-Bandwidth Memory (HBM) to the accelerator's cache. While methods such as speculative decoding have been suggested to address this issue, their implementation is impeded by the challenges associated with acquiring and maintaining a separate draft model. In this paper, we present Medusa, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, Medusa constructs multiple candidate continuations and verifies them simultaneously in each decoding step. By leveraging parallel processing, Medusa substantially reduces the number of decoding steps required. We present two levels of fine-tuning procedures for Medusa to meet the needs of different use cases: Medusa-1: Medusa is directly fine-tuned on top of a frozen backbone LLM, enabling lossless inference acceleration. Medusa-2: Medusa is fine-tuned together with the backbone LLM, enabling better prediction accuracy of Medusa heads and higher speedup but needing a special training recipe that preserves the backbone model's capabilities. Moreover, we propose several extensions that improve or expand the utility of Medusa, including a self-distillation to handle situations where no training data is available and a typical acceptance scheme to boost the acceptance rate while maintaining generation quality. We evaluate Medusa on models of various sizes and training procedures. Our experiments demonstrate that Medusa-1 can achieve over 2.2x speedup without compromising generation quality, while Medusa-2 further improves the speedup to 2.3-3.6x.
hub tools
representative citing papers
BubbleSpec exploits long-tail bubbles in synchronous RL by using faster ranks' idle time to pre-generate rollout drafts for speculative decoding, reducing steps by 50% and raising throughput up to 1.8x while preserving exact synchrony.
FASER delivers up to 53% higher throughput and 1.92x lower latency in dynamic LLM serving by adjusting speculative lengths per request, early pruning of rejects, and overlapping draft/verification phases via frontiers.
NI Sampling accelerates discrete diffusion language models up to 14.3 times by training a neural indicator to select which tokens to sample at each step using a trajectory-preserving objective.
WISV uses a channel-aware semantic acceptance policy on hidden representations to boost accepted sequence length by up to 60.8% and cut interaction rounds by 37.3% in distributed speculative decoding, with under 1% accuracy loss.
Visual token pruning in MLLMs fails on complex reasoning due to Relevant Visual Information Shift during decoding, but the DSTP framework fixes it training-free across models.
Sequential KV compression via probabilistic language tries and predictive delta coding achieves 3.3-4.3 bits per token entropy, yielding up to 914x better ratios than TurboQuant even with large overhead.
Language models trained on parallel streams of computation can overcome single-stream bottlenecks in autonomous agents by enabling simultaneous reading, thinking, and acting.
BitLM replaces per-token softmax with bitwise continuous diffusion inside causal blocks to generate multiple tokens in parallel while preserving autoregressive structure.
ME-DLM augments parallel masked diffusion models with edit-distance-supervised refinements to raise quality on coding and math benchmarks while using far fewer diffusion steps.
PARD-2 uses Confidence-Adaptive Token optimization to align draft model training with acceptance length in speculative decoding, enabling dual-mode operation and up to 6.94x lossless speedup on Llama3.1-8B.
CASCADE formalizes semantic interchangeability and convergence in target model representations to enable context-aware acceptance relaxation in tree-based speculative decoding, delivering up to 3.6x speedup on text-to-image models without quality loss.
WindowQuant performs window-adaptive mixed-precision KV cache quantization guided by similarity to the text prompt, with reordering to enable efficient inference in VLMs.
CoVSpec achieves up to 2.21x higher throughput and over 96% lower communication overhead for device-edge VLM inference via training-free visual token reduction, adaptive drafting, and decoupled parallel verification-correction in speculative decoding.
EVICT adaptively truncates draft trees in MoE speculative decoding by combining drafter signals with profiled costs to retain only cost-effective prefixes, delivering up to 2.35x speedup over autoregressive decoding.
SPIN co-designs sparse attention with hierarchical memory to achieve 1.66-5.66x higher throughput, 7-9x lower TTFT, and up to 58% lower TPOT than vLLM and original sparse implementations.
RACER unifies retrieval of exact matching patterns with logit-driven cues to produce better speculative drafts, achieving more than 2x speedup over autoregressive decoding and outperforming prior training-free speculative decoding methods.
SMART uses marginal benefit-cost analysis to dynamically build efficient speculative trees, achieving 15-20% additional speedup in LLM and MLLM inference.
SnapKV selects clustered important KV positions per attention head from an observation window at the prompt end, yielding 3.6x faster generation and 8.2x better memory efficiency on 16K-token inputs with comparable performance across 16 datasets.
STOP is a new learnable internal path-pruning technique that improves efficiency and accuracy of parallel reasoning in LRMs under fixed compute budgets.
Empirical measurements across four NLP domains show task type is a stronger predictor of speculative decoding acceptance than tree depth, with chat uniquely achieving expected accepted length over 1 token per step.
DMax enables faster parallel decoding in diffusion language models by using on-policy training to recover from errors and soft embedding interpolations for iterative revision, boosting tokens per forward pass roughly 2-3x on benchmarks while preserving accuracy.
A framework combines multi-LoRA runtime switching, multi-stream stylistic decoding, and Dynamic Self-Speculative Decoding with INT4 quantization to achieve 4-6x memory and latency gains for on-device inference of a one-for-all foundational LLM on Qualcomm chipsets.
citing papers explorer
-
WISV: Wireless-Informed Semantic Verification for Distributed Speculative Decoding in Device-Edge LLM Inference
WISV uses a channel-aware semantic acceptance policy on hidden representations to boost accepted sequence length by up to 60.8% and cut interaction rounds by 37.3% in distributed speculative decoding, with under 1% accuracy loss.