High-level synthesis (HLS) is a widely used tool in designing Field Programmable Gate Array (FPGA). HLS enables FPGA design with software programming languages by compiling the source code into an FPGA circuit. The source code includes a program (called ``kernel'') and several pragmas that instruct hardware synthesis, such as parallelization, pipeline, etc. While it is relatively easy for software developers to design the program, it heavily relies on hardware knowledge to design the pragmas, posing a big challenge for software developers. Recently, different machine learning algorithms, such as GNNs, have been proposed to automate the pragma design via performance prediction. However, when applying the trained model on new kernels, the significant domain shift often leads to unsatisfactory performance. We propose a more domain-generalizable model structure: a two-level hierarchical Mixture of Experts (MoE), that can be flexibly adapted to any GNN model. Different expert networks can learn to deal with different regions in the representation space, and they can utilize similar patterns between the old kernels and new kernels. In the low-level MoE, we apply MoE on three natural granularities of a program: node, basic block, and graph. The high-level MoE learns to aggregate the three granularities for the final decision. To stably train the hierarchical MoE, we further propose a two-stage training method. Extensive experiments verify the effectiveness of the hierarchical MoE.

Large language models (LLMs) based on transformer architecture have shown outstanding performance across numerous real-world tasks. However, the autoregressive nature of these models makes the inference process slow and costly. Speculative decoding has emerged as a promising solution, leveraging a smaller auxiliary model to draft future tokens, which are then validated simultaneously by the larger model, achieving a speed-up of 1-2x. Although speculative decoding matches the same distribution as multinomial sampling, multinomial sampling itself is prone to suboptimal outputs, where as beam sampling is widely recognized for producing higher-quality results by maintaining multiple candidate sequences at each step. This paper explores the novel integration of speculative decoding with beam sampling. However, there are four key challenges: (1) how to generate multiple sequences from the larger model's distribution given drafts sequences from the small model; (2) how to dynamically optimize the number of beams to balance efficiency and accuracy; (3) how to efficiently verify the multiple drafts in parallel; and (4) how to address the extra memory costs inherent in beam sampling. To address these challenges, we propose dynamic-width speculative beam decoding (DSBD). Specifically, we first introduce a novel draft and verification scheme that generates multiple sequences following the large model's distribution based on beam sampling trajectories from the small model. Then, we introduce an adaptive mechanism to dynamically tune the number of beams based on the context, optimizing efficiency and effectiveness. Besides, we extend tree-based parallel verification to handle multiple trees simultaneously, accelerating the verification process. Finally, we illustrate a simple modification to our algorithm to mitigate the memory overhead of beam sampling. Experimental results show that our approach achieves a 1.5-1.9x speed-up and1.8-2.5x lower energy consumption compared to beam sampling, with no loss in downstream performance. Moreover, it can produce significantly higher-quality outputs than speculative decoding, while maintaining similar time, memory, and energy costs. In summary, our method offers a more efficient and effective inference process for LLMs.

Large language models (LLMs) have achieved remarkable success across diverse tasks, yet their inference processes are hindered by substantial time and energy demands due to single-token generation at each decoding step. While previous methods such as speculative decoding mitigate these inefficiencies by producing multiple tokens per step, each token is still generated by its single-token distribution, thereby enhancing speed without improving effectiveness. In contrast, our work simultaneously enhances inference speed and improves the output effectiveness. We consider multi-token joint decoding (MTJD), which generates multiple tokens from their joint distribution at each iteration, theoretically reducing perplexity and enhancing task performance. However, MTJD suffers from the high cost of sampling from the joint distribution of multiple tokens. Inspired by speculative decoding, we introduce multi-token assisted decoding (MTAD), a novel framework designed to accelerate MTJD. MTAD leverages a smaller auxiliary model to approximate the joint distribution of a larger model, incorporating a verification mechanism that not only ensures the accuracy of this approximation, but also improves the decoding efficiency over conventional speculative decoding. Theoretically, we demonstrate that MTAD closely approximates exact MTJD with bounded error. Empirical evaluations using Llama-2 and OPT models ranging from 13B to 70B parameters across various tasks reveal that MTAD reduces perplexity by 21.2% and improves downstream performance compared to standard single-token sampling. Furthermore, MTAD achieves a 1.42× speed-up and consumes 1.54× less energy than conventional speculative decoding methods. These results highlight MTAD’s ability to make multi-token joint decoding both effective and efficient, promoting more sustainable and high-performance deployment of LLMs.

Zifan He, Yingqi Cao, Zongyue Qin, Neha Prakriya, Yizhou Sun, Jason Cong. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Speculative decoding is a technique to leverage hardware concurrency in order to enable multiple steps of token generation in a single forward pass, thus improving the efficiency of large-scale autoregressive (AR) Transformer models. State-space models (SSMs) are already more efficient than AR Transformers, since their state summarizes all past data with no need to cache or re-process tokens in the sliding window context. However, their state can also comprise thousands of tokens; so, speculative decoding has recently been extended to SSMs. Existing approaches, however, do not leverage the tree-based verification methods, since current SSMs lack the means to compute a token tree efficiently. We propose the first scalable algorithm to perform tree-based speculative decoding in state-space models (SSMs) and hybrid architectures of SSMs and Transformer layers. We exploit the structure of accumulated state transition matrices to facilitate tree-based speculative decoding with minimal overhead relative to current SSM implementations. Along with the algorithm, we describe a hardware-aware implementation that improves naive application of AR Transformer tree-based speculative decoding methods to SSMs. Furthermore, we outperform vanilla speculative decoding with SSMs even with a baseline drafting model and tree structure on three different benchmarks, opening up opportunities for further speed up with SSM and hybrid model inference. Code can be find at: https://github.com/wyc1997/stree.

There have been several recent works proposed to utilize model-based optimization methods to improve the productivity of using high-level synthesis (HLS) to design domain-specific architectures. They would replace the time-consuming performance estimation or simulation of design with a proxy model, and automatically insert pragmas to guide hardware optimizations. In this work, we address the challenges associated with high-level synthesis (HLS) design space exploration (DSE) through the evolving landscape of HLS tools. As these tools develop, the quality of results (QoR) from synthesis can vary significantly, complicating the maintenance of optimal design strategies across different toolchains. We introduce Active-CEM, a task transfer learning scheme that leverages a model-based explorer designed to adapt efficiently to changes in toolchains. This approach optimizes sample efficiency by identifying high-quality design configurations under a new toolchain without requiring extensive re-evaluation. We further refine our methodology by incorporating toolchain-invariant modeling. This allows us to predict QoR changes more accurately despite shifts in the black-box implementation of the toolchains. Experiment results on the HLSyn benchmark transitioning to new toolchain show an average performance improvement of 2.38× compared to AutoDSE and a 1.2× improvement over HARP, while also increasing the sample efficiency by 5.75×, and reducing the runtime by 2.7×.

High-level synthesis (HLS) aims to raise the abstraction layer in hardware design, enabling the design of domain-specific accelerators (DSAs) like field-programmable gate arrays (FPGAs) using C/C++ instead of hardware description languages (HDLs). Compiler directives in the form of pragmas play a crucial role in modifying the microarchitecture within the HLS framework. However, the space of possible microarchitectures grows exponentially with the number of pragmas. Moreover, the evaluation of each candidate design using the HLS tool consumes significant time, ranging from minutes to hours, leading to a time-consuming optimization process. To accelerate this process, machine learning models have been used to predict design quality in milliseconds. However, existing open-source datasets for training such models are limited in terms of design complexity and available optimizations. In this paper, we present HLSyn, the first benchmark that addresses these limitations. It contains more complex programs with a wider range of optimization pragmas, making it a comprehensive dataset for training and evaluating design quality prediction models. The HLSyn benchmark consists of 42 unique programs/kernels, resulting in over 42,000 labeled designs. We conduct an extensive comparison of state-of-the-art baselines to assess their effectiveness in predicting design quality. As an ongoing project, we anticipate expanding the HLSyn benchmark in terms of both quantity and variety of programs to further support the development of this field.

Graph similarity search aims to find the most similar graphs to a query in a graph database in terms of a given proximity measure, say Graph Edit Distance (GED). It is a widely studied yet still challenging problem. Most of the studies are based on the pruning-verification framework, which first prunes non-promising graphs and then conducts verification on the small candidate set. Existing methods are capable of managing databases with thousands or tens of thousands of graphs, but fail to scale to even larger database, due to their exact pruning strategy. Inspired by the recent success of deep-learning-based semantic hashing in image and document retrieval, we propose a novel graph neural network (GNN) based semantic hashing, i.e. GHashing, for approximate pruning. We first train a GNN with ground-truth GED results so that it learns to generate embeddings and hash codes that preserve GED between graphs. Then a hash index is built to enable graph lookup in constant time. To answer a query, we use the hash codes and the continuous embeddings as two-level pruning to retrieve the most promising candidates, which are sent to the exact solver for final verification. Due to the approximate pruning strategy leveraged by our graph hashing technique, our approach achieves significantly faster query time compared to state-of-the-art methods while maintaining a high recall. Experiments show that our approach is on average 20x faster than the only baseline that works on million-scale databases, which demonstrates GHashing successfully provides a new direction in addressing graph search problem for large-scale graph databases.