analiza aiqbench
SpecQuant introduces a training-free framework that dynamically routes LLM inference between multiple quantized variants based on task complexity, achieving significant speedups without retraining. This approach addresses the critical bottleneck of running large models on consumer hardware by optimizing compute and memory usage through adaptive precision switching. The method offers a practical pathway for efficient local deployment across diverse devices while maintaining high accuracy.
SpecQuant: Speculative Decoding with Multi-Parent Quantization for Adaptive LLM Inference
arXiv:2609.21704v1 Announce Type: new Abstract: Running large language models (LLMs) locally continues to be limited by restrictions of compute and memory on consumer hardware. The popular acceleration technologies, such as quantization, speculative decoding, and adaptive inferencing, offer substantial speed boosts but usually necessitate retraining, per architecture tuning, or draft models. SpecQuant is a trainingfree framework, that combines speculative decoding with multiparent quantization to perform adaptive, efficient inference of LLMs. SpecQuant derives multiple quantized variants (INT4, FP8, FP16) from a shared base model, and dynamically routes queries based on predicted complexity; lightweight variants are used for simple or factual tasks, and full-precision models are used for complex reasoning tasks or long-context inputs. The shared-weight design of SpecQuant ensures sufficient token acceptance for speculative decoding without compatibility issues using separate draft parent models. We evaluate SpecQuant on Qwen2.5 based models on the MMLU, AlpacaEval, and GSM8K datasets, or benchmarks, demonstrating 35-43% speedups without degrading accuracy greater than 2%, substantial within the LLM community. SpecQuant enables practical on-device LLM deployment across diverse hardware without special infrastructure or expertise.
Abstract:Running large language models (LLMs) locally continues to be limited by restrictions of compute and memory on consumer hardware. The popular acceleration technologies, such as quantization, speculative decoding, and adaptive inferencing, offer substantial speed boosts but usually necessitate retraining, per architecture tuning, or draft models. SpecQuant is a trainingfree framework, that combines speculative decoding with multiparent quantization to perform adaptive, efficient inference of LLMs. SpecQuant derives multiple quantized variants (INT4, FP8, FP16) from a shared base model, and dynamically routes queries based on predicted complexity; lightweight variants are used for simple or factual tasks, and full-precision models are used for complex reasoning tasks or long-context inputs. The shared-weight design of SpecQuant ensures sufficient token acceptance for speculative decoding without compatibility issues using separate draft parent models. We evaluate SpecQuant on Qwen2.5 based models on the MMLU, AlpacaEval, and GSM8K datasets, or benchmarks, demonstrating 35-43% speedups without degrading accuracy greater than 2%, substantial within the LLM community. SpecQuant enables practical on-device LLM deployment across diverse hardware without special infrastructure or expertise.
5 pages, 1 figure. Published in the 2026 Fifth International Conference on Power, Control and Computing Technologies (ICPC2T)
Machine Learning (cs.LG)
arXiv:2609.21704 [cs.LG]
(or arXiv:2609.21704v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2609.21704 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
2026 Fifth International Conference on Power, Control and Computing Technologies (ICPC2T), Raipur, India, 11-13 March 2026, pp. 371-375, IEEE, 2026
https://doi.org/10.1109/ICPC2T68221.2026.11646348 Focus to learn more DOI(s) linking to related resources
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 312 cuvinte
Subiecte conexe
Optimizing ATOM and vLLM-ATOM for High-Interactivity Inference
AMD · 1 sept. 2026
Amazon SageMaker Inference: 2026 year-to-date launches in review
AWS · 18 sept. 2026
[AINews] High Taste Testers Ep. 1
Latent Space · 17 oct. 2026
ATOM: Unlocking Extreme AMD Instinct Inference with Software-Hardware Co-Optimization
AMD · 15 iun. 2026