This paper presents FastTPS, a high performance and low-precision loss method for accelerating the token-phase in LLM inference on general AI accelerators which includes three key components: AI accelerator-enabled reloading-free KV Cache concatenation which decreases memory access overhead as well as enables full fusion of Attention.
Abstract
The popularity of large language models (LLMs) escalates an ongoing demand for effective inference. However, due to the sequential processing of tokens during the token phase in decoder-only LLMs inference, the inherent low parallelism leads to reduced throughput and suboptimal utilization of the computing units on artificial intelligence (AI) accelerators, particularly when handling long-sequence inputs that impose significant memory overhead. Recently, many reported methods have been developed as potential solutions, since they emerge with numeric deviation. This paper presents FastTPS, a high performance and low-precision loss method for accelerating the token-phase in LLM inference on general AI accelerators which includes three key components: (1) AI accelerator-enabled reloading-free KV Cache concatenation which decreases memory access overhead as well as enables full fusion of Attention, (2) high-efficiency and high-accuracy'RoPE'attention based on the tiling optimized FLAT, and (3) highly-fused MLP with fine-grain pipeline scheduling. Our results confirm that FastTPS significantly alleviates memory bottlenecks in the token phase, delivering a 6x speed improvement (compared to none-fusion) on an AMD Ryzen AI 300 series NPU with BF16 precision while sustaining 93% peak memory bandwidth utilization during Phi3-mini-4k-instruct inference.
Mamba has emerged as a promising alternative to Transformers for on-device LLM inference, offering linear complexity and eliminating KV-cache. However, autoregressive decoding reloads full model weights every token, accounting for 93.2% of total inference energy, with no existing accelerator resolving this overhead. We present HotBa, a heterogeneous Mamba accelerator that reduces per-token weight transfer by 82% and redundant computation by 59% via Δ-guided early rejection for wide-tree speculative decoding in Mamba, while a heterogeneous INT8/FP16 core and tree management unit achieve 40.4× area efficiency and 5.18× SSM speedup with 0.4% area overhead. Synthesized in 28nm CMOS, HotBa achieves 75.32 tokens/s with 1.65× speedup and 7.55× energy efficiency over the state-of-the-art Mamba accelerator.
Ghangmin Yun, Jueun Jung, Bokyoung Seo et al.· Proceedings of the ACM/IEEE...· 0 citations
This paper presents FlashAttention-V, a blocked FlashAttention for scalable vector architectures that adapts efficiently from short to very long vectors by exploiting parallelism across attention heads, inter-head packing to enable efficient utilization of vector lengths beyond the head dimension, and improving vector register utilization and memory access locality.
Prefill or prompt processing and Decode or token generation are two distinct subphases of LLM inference that are greatly influenced by LLM accelerators such as GPT-Generated Unified Format (GGUF Q4_K_M), NormalFloat 4-bit (NF4) Quantization, FlashAttention-2 and others. Although these accelerators clearly improve end-to-end LLM inference performance, their effectiveness over these subphases remains largely understudied. To address this gap, we present a cross-platform, multi-model empirical study, where we deploy multiple ∼ 1B-parameter LLMs on GPU, CPU, and Raspberry Pi 4B edge hardware platforms in the presence and absence of these accelerators. Each test case evaluates 10,000+ inference runs with separate phase-wise and end-to-end performance indicators. Our study brings several important observations, including the contrastive effect of quantization under different hardware bottlenecks, along with a quantification of runtime delays (up to +139%) caused by the lack of parallelism in the ARM architecture. Based on these benchmarking results and observations, we identify several open research challenges in the concluding section. Our work is fully reproducible and open-sourced on GitHub1.
Subhransu Das, Jiaming Cheng, Swathi Vallabhajosyula et al.· Practice and Experience in A...· 0 citations
LLM inference is increasingly constrained by accelerator memory capacity rather than compute throughput. This constraint is especially acute in single-accelerator and small-node inference systems, where limited on-package memory capacity restricts the size of deployable models. HBF is an emerging 3D-stacked NAND flash technology that provides multi-terabyte near-accelerator capacity, making it a promising capacity tier for storing LLM weights. However, existing HBF-based proposals face three adoption challenges: they (1) rely on coarse-grained static prefetching for LLM weights aiming to hide the microsecond-level read latency of the NAND flash device while maximizing HBF's read throughput, (2) expose NAND flash management tasks (e.g., refresh operations) to the accelerator-visible critical inference path, and (3) miss optimization opportunities to specialize and optimize the flash-management mechanisms to the workload behavior. Our goal is to design an efficient HBF substrate that integrates HBF as a memory-capacity tier alongside HBM while addressing these three challenges. To this end, we propose FLINT, a workload-driven HBF substrate for capacity-scalable LLM inference. FLINT introduces three mechanisms: (1) a hardware burst-buffer controller that dynamically coalesces and pipelines HBF reads aiming to utilize existing NAND flash buffers while sustaining high HBF bandwidth, (2) a phantom-plane refresh mechanism, which removes refresh from the critical inference path by moving refresh-related NAND flash operations outside the read foreground back via low-cost resource duplication, and (3) a read-only FTL, which replaces SSD-class support for arbitrary writes with a compact table that translates logical weight bursts to physical HBF locations.
Geraldo F. Oliveira, Arash Tavakkol, Xiang-Yu Zhu et al.· 0 citations
On-device deployment of Large Language Models (LLMs) has become essential for personalized edge applications. A primary bottleneck is external memory access (EMA) in feed-forward network (FFN) layers. Speculative decoding and mixture-of-experts (MoE) are promising solutions. Speculative decoding reduces the number of decoding stages by generating multiple tokens per stage, and MoE minimizes per-stage cost through sparse expert activation. However, there is an incompatibility when combining these two techniques. We propose EdgeXpert, a software-hardware co-designed LLM accelerator that resolves this incompatibility. In the prefill stage, the prompt-wise expert reuse reformulates routing as prompt-level expert reuse rather than independent per-token expert selection. It identifies important tokens using a lightweight encoder, constructs a shared expert set from them, and routes less important tokens with a reduced expert budget to lower expert EMA. In the decode stage, depth-aware expert coalescing exploits the contextual similarity and mutual exclusivity of same-depth candidate tokens. Rather than loading the union of all required channels, EdgeXpert loads only salient channels and applies computational calibration to recover accuracy without additional memory access. Synthesized in Samsung 28nm technology at 800 MHz, EdgeXpert achieves up to 56.3% latency reduction and 44.1% energy reduction compared to prior works, while maintaining near-baseline accuracy.
Sangwoo Ha, Hyunwoo Seo, Y. Jo et al.· 0 citations
The impressive performance of large language models (LLMs) in various artificial intelligence tasks has not yet been applied to real-time signal processing tasks due to the prohibitive memory and computation requirements. The paper outlines an efficient scheme of inference in LLM by synergistically using post-training quantization, key-value (KV) cache compression, speculative decoding, and Flash Attention. We look at the combined effect that these techniques have on reducing memory footprint by as much as 4x and enhancing decoding by up to 3x without drastically compromising the accuracy of signal interpretation. The results of experimental assessments on spectrum sensing, modulation classification, and channel estimation tasks indicate that our optimized pipeline can reach almost the baseline performance, and it can be implemented on the resource-constrained hardware accelerators, such as GPUs, TPUs, and edge neural processing units. The suggested methodology will fill the gap between the state-of-the-art LLM capabilities and the latency constraints of the intelligent signal processing systems, paving a viable way to LLM-based cognition in the next-generation wireless and IoT networks.
Feng Zhou· International Conference on...· 0 citations