Mix-Quant: Quantized Prefilling, Precise Decoding for Agentic LLMs

arXiv:2605.20315 · 2026-05-21 공개 · arXiv · PDF

llm-agents long-context quantization nvfp4 prefilling agentic-inference efficiency bf16

Abstract

LLM agents have recently emerged as a powerful paradigm for solving complex tasks through planning, tool use, memory retrieval, and multi-step interaction. However, these agentic workflows often introduce substantial input-side overhead, making the compute-intensive prefilling stage a key bottleneck in long-context, multi-turn inference. In this work, we propose Mix-Quant, a simple and effective phase-aware quantization framework for fast agentic inference. We first investigate FP4 quantization in agentic LLM workflows and observe that quantizing the entire inference process can incur significant performance degradation. In contrast, the prefilling stage exhibits substantial quantization redundancy and can therefore be quantized with minimal accuracy loss, despite being the dominant source of computation. Based on this insight, we apply high-throughput NVFP4 quantization to the prefilling phase while preserving BF16 precision for decoding. By decoupling prefilling acceleration from decoding quality, Mix-Quant combines phase-aware algorithmic quantization with hardware-efficient NVFP4 execution to alleviate the inference bottleneck in LLM agents. Extensive experiments across long-context and agentic benchmarks demonstrate that Mix-Quant largely preserves task performance while delivering significant efficiency improvements, achieving up to a 3x speedup during prefilling.

한국어 요약

📋 한 줄 요약

**[LLM 추론 효율화 / 양자화]** 에이전트 LLM 워크플로우의 prefilling은 공격적으로 양자화하고 decoding은 BF16으로 유지하는 phase-aware 양자화 프레임워크 Mix-Quant 제안.

🎯 핵심 기여도

💡 핵심 아이디어

LLM 에이전트는 prefilling이 연산 비중을 지배하지만 동시에 더 양자화에 견고하므로, "비싼 단계는 공격적으로 양자화하고 민감한 단계는 보존"하는 단계별 차등 양자화로 효율-품질 트레이드오프를 깨뜨릴 수 있다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: 에이전트 LLM의 실제 병목(prefilling)을 직접 겨냥하는 phase-aware 추론 가속 패러다임의 실증, NVFP4 시대의 실무적 처방 제공. **한계**: 평가가 NVFP4 가능한 최신 하드웨어에 종속, decoding이 길어지는 도메인에서는 가속 이득이 제한될 수 있음.

🚀 실용적 활용