Towards Large Model Feature Coding

2026-05-20

AI Summary

This work addresses the challenges of deploying large language models under constraints of computation, memory, and privacy. While model splitting alleviates resource demands, it introduces significant bottlenecks in the efficient transmission and storage of heterogeneous intermediate features. To tackle this, the paper introduces Large Model Feature Coding (LaMoFC) as a foundational system component and presents LaMoFCBench—the first systematic benchmarking framework encompassing diverse tasks, architectures, and splitting scenarios. The framework defines representative split points and collects heterogeneous features across multiple granularities, modalities, and autoregressive context caches. Through a unified evaluation pipeline, the study comprehensively reveals a fundamental mismatch between existing general-purpose codecs and the unique characteristics of large-model features, thereby establishing a reproducible empirical foundation to advance next-generation feature compression techniques.

Abstract

Large models have delivered remarkable performance across a wide range of perception and generation tasks, yet practical deployment is increasingly constrained by computational and memory budgets, as well as privacy requirements. Split execution alleviates these constraints by partitioning computation across devices, but it inevitably introduces intensive transmission and storage of intermediate features. Unlike conventional feature coding for CNNs that typically targets homogeneous spatial activation maps, modern large models generate heterogeneous features with varying statistical distributions and compression tolerances, e.g., multi-level/multi-modal representations and autoregressive context caches. These characteristics necessitate treating large model feature coding (LaMoFC) as a fundamental system component and call for a systematic evaluation framework. In this paper, we present a comprehensive benchmark and evaluation framework for LaMoFC. We first build the feature dataset LaMoFCBench, covering diverse task requirements across 4 categories and 16 scenarios while integrating widelyadopted architectures and various split-computing settings. We then specify representative split points according to practical application scenarios to extract intermediate features, establishing a unified pipeline for fair and reproducible comparisons. Finally, we benchmark mainstream universal feature codecs, exposing the profound misalignment between existing coding paradigms and the heterogeneous nature of large model features. These findings reveal that LaMoFC demands a fundamental departure from existing paradigms, and LaMoFCBench provides the shared empirical foundation to drive this transition. The data and code will be available at https://github.com/lartpang/LaMoFCBench.

Problem

• large model
• feature coding
• split computing
• heterogeneous features
• intermediate feature compression

Innovation

• Large Model Feature Coding
• Split Computing
• Heterogeneous Features
• Feature Compression
• Benchmarking Framework