aiqbench analysis
This research proposes a novel Compute-in-Flash architecture that leverages integer-only quantization and sparse dictionary coding to accelerate LLM inference on SSD-based memory devices. By compressing KV caches and eliminating floating-point operations, the method significantly reduces dynamic traffic while maintaining model accuracy for major open-source weights like Llama-3.1 and Qwen-2.5. The work directly addresses critical hardware bottlenecks in scaling long-context applications, offering a viable path toward cost-effective inference at the memory level.
LLM Inference in a Flash!
arXiv:2609.16161v1 Announce Type: new Abstract: Large Language Models (LLMs) have shown impressive capabilities across a range of natural language processing tasks, and LLM inference has emerged as a critical workload for enabling downstream applications. The demands of serving LLM inference are becoming increasingly challenging as requests shift toward longer sequences and heavier inference, driven by retrieval-augmented generation, inference-time compute scaling, and long-context applications. Additionally, these challenges are compounded by hardware trends, as memory capacity and communication bandwidth are not scaling as fast as increases in workload complexity. Compute-in-Flash is a promising solution to address memory bandwidth limitations by moving computation close to memory, and to exploit the large capacity of SSD technologies. However, it is challenging to deploy LLMs on these systems as they lack support for high-precision floating point operations and have limited write endurance. In our work, we aim to address these challenges by designing inference algorithms to enable LLM inference on Flash compute-in-memory devices. We present an end-to-end integer-only quantization approach to eliminate expensive floating-point computations. To address the limited write endurance, we design a dictionary-based KV cache compression strategy based on sparse dictionary coding that represents each KV vector as a linear combination of static dictionary vectors. These algorithmic improvements enable us to exploit the benefits of Compute-in-Flash for both model weights and KV cache, and to minimize expensive data transfer operations. Across Llama-3.1-8B and Qwen-2.5-7B, our combined method exhibits limited accuracy degradation while reducing dynamic KV cache traffic by 15$\times$.
Title:LLM Inference in a Flash!
Abstract:Large Language Models (LLMs) have shown impressive capabilities across a range of natural language processing tasks, and LLM inference has emerged as a critical workload for enabling downstream applications. The demands of serving LLM inference are becoming increasingly challenging as requests shift toward longer sequences and heavier inference, driven by retrieval-augmented generation, inference-time compute scaling, and long-context applications. Additionally, these challenges are compounded by hardware trends, as memory capacity and communication bandwidth are not scaling as fast as increases in workload complexity. Compute-in-Flash is a promising solution to address memory bandwidth limitations by moving computation close to memory, and to exploit the large capacity of SSD technologies. However, it is challenging to deploy LLMs on these systems as they lack support for high-precision floating point operations and have limited write endurance. In our work, we aim to address these challenges by designing inference algorithms to enable LLM inference on Flash compute-in-memory devices. We present an end-to-end integer-only quantization approach to eliminate expensive floating-point computations. To address the limited write endurance, we design a dictionary-based KV cache compression strategy based on sparse dictionary coding that represents each KV vector as a linear combination of static dictionary vectors. These algorithmic improvements enable us to exploit the benefits of Compute-in-Flash for both model weights and KV cache, and to minimize expensive data transfer operations. Across Llama-3.1-8B and Qwen-2.5-7B, our combined method exhibits limited accuracy degradation while reducing dynamic KV cache traffic by 15$\times$.
Machine Learning (cs.LG)
arXiv:2609.16161 [cs.LG]
(or arXiv:2609.16161v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2609.16161 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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.
Full text under the publisher's open licence · 339 words
Related coverage
[AINews] DeepSeek v4.1-Flash: 763B-P8B-D16B novel causal Encoder–Decoder architecture with vision marks the Return of the Whale
Latent Space · Sep 12, 2026
[AINews] High Taste Testers Ep. 1
Latent Space · Oct 17, 2026
[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
Latent Space · Oct 13, 2026
[AINews] not much happened today
Latent Space · Sep 10, 2026