Searcharxiv⌕ Search

arXiv subjects

Weng-Fai Wong

Publications and source records attributed to Weng-Fai Wong.

At least 19 recordsLinked to original sources

MHRC-Bench: A Multilingual Hardware Repository-Level Code Completion benchmark

Large language models (LLMs) have achieved strong performance on code completion in general-purpose programming languages, but existing repository-level benchmarks focus almost exclusively on software code and largely overlook hardware description languages (HDLs). We present \textbf{MHRC-Bench}, consisting of \textbf{MHRC-Bench-Train} and \textbf{MHRC-Bench-Eval}, the first benchmark for multilingual hardware code completion at the repository level with functional evaluation. MHRC-Bench covers three major hardware design coding styles and annotates each completion target with code-structure-level and hardware-oriented semantic labels derived from concrete syntax tree (CST) analysis. Post-training billion-scale code LLMs on MHRC-Bench-Train enables smaller open-weight models to outperform much larger general-purpose models under matched context windows. Beyond aggregate accuracy, our analysis surfaces four findings that distinguish hardware from software code completion: \emph{(i)}~performance varies sharply across HDLs, with high-level synthesis (HLS) easiest and VHDL hardest in most settings across functional pass rate, exact match (EM), and edit similarity (ES); \emph{(ii)}~hardware accuracy is non-monotonic in code-structure depth, in contrast with the monotonic depth--accuracy curve known for software; \emph{(iii)}~accuracy plateaus once the input context exceeds 2{,}048 tokens, while software completion continues to benefit from longer context; and \emph{(iv)}~certain repository-level retrievers tuned on software underperform a no-retrieval baseline on HDL repositories because their structural and dependency assumptions are violated by hardware code.

cs.PL↗

Reward-Weighted On-Policy Distillation with an Open Property-Equivalence Verifier for NL-to-SVA Generation

Specialized language models for translating natural language (NL) into SystemVerilog assertions (SVA) reach approximately 76\% accuracy on NL2SVA-Human, but aggregate scores obscure errors on bounded-delay and liveness specifications. Models often reuse a small set of implication templates where different temporal behavior is required. Supervised fine-tuning rewards agreement with reference tokens, whereas SVA correctness depends on property equivalence. We propose \emph{reward-weighted on-policy distillation} (RWOPD), which uses an open SymbiYosys+Z3 property-equivalence checker to select student rollouts and weight their distillation loss. A frozen 14B teacher supplies forward Kullback--Leibler (KL) supervision at each response token; equivalent rollouts receive greater weight than one-sided implications. Starting from curriculum supervised fine-tuning (SFT), RWOPD distills CodeV-SVA-14B into a Qwen2.5-Coder-7B-Instruct student. The student has the highest reported pass@1, pass@5, and pass@10 on NL2SVA-Human and NL2SVA-Machine among the compared models, including specialized baselines and 671B general-purpose models.

cs.AR↗

Rethinking KV Caching for LLM Decode: Free-Space Optical KV Regeneration for Reduction of Memory Traffic

Long-context large language model (LLM) decoding repeatedly reads the historical key-value (KV) cache, making multi-head attention (MHA) memory-bandwidth-bound. Caching layer-input activations instead halves historical-cache storage and traffic, but requires regenerating KV from each cached activation. On digital accelerators such as GPUs, this regeneration takes far longer than fetching the activation from HBM. The KV regeneration overhead outweighs the savings from reduced memory traffic, shifting decode attention from a memory bottleneck to a compute bottleneck. We present OptiDecode, a hybrid optoelectronic decode-attention accelerator that regenerates historical KV using free-space optics and computes attention digitally. Projection weights remain resident on optical tiles, exploiting reuse across historical tokens for high-throughput optical regeneration. Scaling to large models introduces two challenges. First, distributing weights across tiles requires sending the same activations to every head group, while each readout die produces only partial KV. Activation replay shares one set of electrical-to-optical converters across head groups without costly broadcast. In-situ attention avoids transferring partial KV from readout dies to reducers by computing on them locally and exchanging only partial scores, probabilities, and partial outputs. Second, processing an activation chunk through all dependent stages takes longer than one activation-fetch interval. Chunk pipelining overlaps successive chunks so that OptiDecode sustains the HBM activation-fetch rate. OptiDecode demonstrates 1.92-1.99x decode-attention speedup over an ideal GPU KV-read baseline at equal peak HBM bandwidth (A100-B200), for 32K-token contexts across evaluated 30B-175B MHA models, approaching the 2x upper bound from halving historical memory traffic.

cs.AR↗

Skill$^{\infty}$: Recursive Self-Improvement for Agent Skill Optimization

Recursive self-improvement (RSI), in which a system also improves the way it improves, is a long-standing aspiration on the path to AGI. Language agents increasingly store procedural knowledge in reusable skills. Optimizing skills offers a direct way to achieve RSI, because the optimization procedure itself can be encoded as a skill. Existing skill optimizers, however, follow a hand-designed procedure. Even methods that evolve a meta-skill confine it to a fixed pipeline and relearn it for each task. We present Skill$^{\infty}$, which makes skill optimization recursively self-improving: a meta agent follows a meta-skill to optimize task skills and optimizes the meta-skill itself using the same procedure. The optimization is agent-led. The meta agent diagnoses execution traces, proposes candidates, requests independent reviews, and allocates the search budget. The meta-skill defines these steps, allowing self-improvement to rewrite them. This meta-skill is also decoupled from task skills, refined on multiple skill optimization tasks and then reused to optimize task skills across diverse domains. As in deployment, skills at both levels are standard packages loaded through native skill discovery. On 83 SkillsBench tasks, Skill$^{\infty}$ reaches a pass rate of 60.7%, compared with 56.4% for human-curated skills and 54.5% for SkillEvolver.

cs.AI↗

ECO: Energy-Oriented Configuration Optimization for Attention FFN Disaggregated LLM Serving

Energy-efficient LLM serving requires minimizing serving GPU energy while meeting latency and throughput service-level objectives (SLOs). Attention--FFN disaggregation (AFD) enables separate resource allocation and operating controls for attention and expert computation, but their energy effects remain coupled through the execution pipeline. Realizing its energy-saving potential therefore requires navigating a hierarchical configuration space in which deployment structures constrain admissible controls and shape their end-to-end effects. Finding low-energy configurations that meet SLOs is challenging because physical evaluations are costly and only a small fraction of candidates can be measured. We present Energy-Oriented Configuration Optimization (ECO), which jointly searches deployment structures and their admissible operating controls under a limited measurement budget. ECO constructs a structure-aware energy prior from calibrated stage behavior and pipeline dependencies, then learns residual prediction errors with a Gaussian process. Its cost-aware constrained Bayesian optimization prioritizes measurements according to expected energy improvement while accounting for SLO feasibility, execution success, and evaluation cost, and returns the lowest-energy measured feasible configuration. Across all 16 scenarios on A6000 and A100 with Qwen and DeepSeek, ECO's frozen configurations, evaluated on disjoint requests, reduce serving energy by 40.5\% and increase output token rate by 20.7\% on average relative to baselines while meeting target SLOs. Across the 8 A6000 scenarios, its selected feasible energy averages 33.1\% below generic constrained Bayesian optimization and 25.8\% below genetic search.

cs.AR↗

ActTune: Action-Aware Precision and GPU Operating-Point Adaptation for Energy-Efficient Vision-Language-Action Inference

Vision-language-action (VLA) policies repeatedly invoke inference to control robots, making graphics processing unit (GPU) energy a recurring cost of task execution. Reducing energy per inference call, however, may not reduce energy per successful task if numerical errors increase failures or slower inference prolongs execution. We therefore target GPU energy per successful task while preserving task success and keeping the inference-latency increase within 10\%. Our approach builds on two observations: quantization sensitivity varies across action classes, model layers, and weights versus activations; and numerical precision changes the workload, shifting favorable GPU operating points. We introduce ActTune, an action-aware framework that connects layer-wise precision allocation with workload-dependent GPU operating-point selection over requested frequency--power-cap pairs. A lightweight decision tree learns its splits and leaf precision configurations directly from configuration action errors, then selects precision before each policy call. The controller forecasts the next workload and applies the selected GPU operating point asynchronously using a lookup table calibrated under a latency budget. A shared resident quantized weight bank enables configuration switching without weight reconstruction or additional policy evaluations. On LIBERO, a benchmark for lifelong robot learning, ActTune improves mean task success by up to 2.3\% relative to state of the art. Relative to the original BF16 implementations, it delivers up to $2.02\times$ faster inference and, with GPU operating-point adaptation, reduces energy per successful task by up to 76.8\%.

cs.RO↗

Latency and accuracy tradeoffs in Spiking Neural Networks

Spiking neural networks are attractive for low-power speech command recognition, yet their latency has received far less attention than their energy efficiency, and their multi-timestep execution is widely assumed to make them slower than quantized neural networks. This paper challenges the assumption that more local timesteps necessarily imply higher network latency. By overlapping computation across adjacent layers at the timestep level, SNNs may complete execution in less time than comparable bit-serial QNNs. However, this overlap relies on spikes firing on incomplete inputs, and a spike once generated cannot be withdrawn, so its error persists and reduces accuracy. Waiting for more input before firing would seem to improve accuracy at the cost of reduced overlap. Yet we find and prove that this intuition fails at some layers, where even a small increase in waiting can change spike timing and downstream computation, making the network both slower and less accurate. We therefore propose a Pipeline Delay Search method which selects each layer's delay by balancing task-level accuracy gains against added network latency. We then adapt the selected configurations through spike-based quantization-aware training and bounded tuning of firing thresholds and initial membrane potentials. Together, these steps form Falcon, a framework for Fine-grained Analysis of Latency and Controlled firing which systematically analyzes and optimizes SNN latency under a spatial analog compute-in-memory mapping with shared digital engines. We evaluate Falcon on GSCV2 and SSC, achieving competitive accuracies of 96.31 and 83.02 at modeled network-core latencies of 119.64 and 124.00us, respectively. Together, our analysis and results show that SNNs can compute more yet finish faster, and wait longer yet predict worse, highlighting why Falcon matters for both latency and accuracy.

cs.LG↗

Lapis: Laplacian Spiking Attention via First-Spike Timing and Membrane Leakage

Self-attention has become central to spiking vision transformers, yet its query-key scoring is still largely inherited from dense networks. Existing spiking variants either simplify dot product scoring or replace it with discrete operators, but spike timing, the native variable of a spiking network, does not directly define how tokens are related. We propose Lapis, a spiking attention mechanism that scores each token pair by the L1 distance between its query and key first-spike latency vectors under time-to-first-spike coding, and maps this distance to an affinity through a Laplacian kernel. The kernel's exponential decay matches the impulse response of a leaky integrate-and-fire membrane, so the accumulated latency difference determines the decay of a membrane trace, while row normalization reduces to a bit shift under power-of-two rounding. Scoring therefore needs only subtraction, absolute value, and accumulation, and removes all multiplication between query and key channels. Under a matched backbone and training schedule, Lapis reaches 96.56% top-1 accuracy on CIFAR-10, within 0.53 points of dot-product scoring. On ImageNet-1K, it reduces the estimated arithmetic energy of the attention path by 14.5x relative to dense dot-product attention. The deployed 6-bit model attains 83.25% top-1 accuracy at an estimated arithmetic energy of 3.28mJ per image.

cs.NE↗

Reconsidering the Energy Efficiency of Spiking Neural Networks Inference from Analytical Perspectives

Spiking Neural Networks (SNNs) promise higher energy efficiency over conventional Quantized Artificial Neural Networks (QNNs) due to their event-driven, spike-based computation. However, prevailing energy evaluations often oversimplify, focusing on computational aspects while neglecting critical overheads like comprehensive data movements and memory accesses. Such simplifications can lead to misleading conclusions regarding the true energy benefits of SNNs. This paper presents a rigorous re-evaluation. We establish a fair baseline by mapping rate-encoded SNNs with $T$ timesteps to capacity-matched QNNs with $\lceil \log_2(T+1) \rceil$ bits. This ensures both models have comparable representational capacities, as well as similar hardware requirements, enabling meaningful energy comparisons. We introduce a detailed analytical energy model encompassing core computation and data movements. Using this model, we systematically explore a wide parameter space, including intrinsic network characteristics (SNN time window size, spike rate, QNN sparsity, model size, weight bit-level) and hardware characteristics (memory system and network-on-chip). Our analysis identifies specific operational regimes where SNNs genuinely offer superior energy efficiency. For example, under typical neuromorphic hardware conditions, SNNs with moderate time windows ($T = 5$) require an average spike rate ($s_r$) below 5.7% to outperform equivalent QNNs These insights guide the design of truly energy-efficient neural network solutions.

cs.NE↗

Matterhorn: Masked Time-to-First-Spike Encoding by Reassigning the Silent State for Sparse and Energy-Efficient Spiking Transformers

Spiking neural networks (SNNs) promise energy-efficient inference for large language models (LLMs), yet most reported savings rely on compute-operation counts that overlook data movement. Energy characterization of representative spiking transformers on a commercial 22-nm process shows that accumulation contributes less than 3% of total energy, while spike-triggered inter-core transfers and weight reads dominate the cost. This makes time-to-first-spike (TTFS) encoding a natural choice, as it limits each neuron to at most one spike. However, standard TTFS maps the silent state, an all-zero spike train that transmits no events, to the rarely occurring smallest value, while the most common activations still spike. This raises a simple question: why reserve the only cost-free codeword for a rare value? This choice inverts a basic principle of energy-aware coding, under which the zero-event codeword should represent the most common value, rather than a rare extreme. Thus, we introduce masked time-to-first-spike encoding (M-TTFS), which uses a temporal mask to reassign the silent state to the most common activation value, and a dead-zone extension that trades a controlled amount of information for greater sparsity. Built on M-TTFS with dead-zone radius $k{=}1$, our spiking transformer Matterhorn reaches an overall spike rate of 1.64% on GLUE at an average score of 84.64, exceeding the best prior spiking transformer by 1.42 percentage points while consuming 67% less energy, with consistent gains on spiking LLaMA models from 7B to 70B parameters. Together, these results show that under hardware-faithful accounting, the energy advantage of SNNs is not a given: it is earned by encodings that align spikes with the data distribution.

cs.LG↗

Efficient and Robust Spiking Neural Networks for sEMG-Based Muscle Fatigue Detection

Detecting muscle fatigue via surface electromyography (sEMG) is essential for applications in sports, rehabilitation, and wearable health monitoring. Accurate and timely detection of fatigue is crucial for preventing injuries, optimizing physical performance, and ensuring user safety during prolonged activity. However, existing deep learning models are often unsuitable for this task due to their high computational cost and dependence on large-scale data. In this work, we propose an energy-efficient framework for muscle fatigue detection based on Spiking Neural Networks (SNNs), which exploit sparse, event-driven computation and temporal modeling. We further introduce a quantization-compatible training scheme (SDH) that combines multiple regularization terms to improve robustness under noisy conditions. Evaluated on two public sEMG datasets against a broad set of baselines and under seven noise conditions including physically motivated perturbations, our quantized SNNs match or exceed strong baselines while remaining more stable under diverse noise and reducing estimated energy consumption by up to 201.77x. These results demonstrate the framework's strong potential for real-time deployment in low-power wearable systems.

cs.NE↗

Approaching Shannon Bound with Lossless LLM Weight Compression

Large language models (LLMs) now scale to trillions of parameters, driving weight storage into the terabyte regime and creating an acute mismatch with GPU memory capacity. Although lossless compression is widely effective in other domains, it remains underutilized in LLM systems. Through a comprehensive entropy study across models from 1.5B to 405B parameters and numeric formats ranging from bf16 to int4 and AWQ/SQ8, we find that LLM weights contain far less intrinsic randomness than their stored bitwidth implies, their effective entropy is 2-10x lower, indicating that up to a 10x footprint reduction is theoretically achievable without altering any weight values. Leveraging this insight, we introduce a tile-level, on-the-fly lossless decompression framework based on Asymmetric Numeral Systems that aligns decoding with the GEMM tiling pattern of GPU inference. Our design achieves bit-rates within 0.01-0.1 bits of the Shannon limit across a wide range of LLM numerical formats, demonstrating that nearly all statistical redundancy is eliminated. Integrated into the SGLang serving framework with multi-GPU support, our approach increases the maximum batch size of Qwen-14B from 47 to 75, improving throughput by up to 1.2x. On Mixtral-176B, the feasible batch size increases from 20 to 95 (4.8x), yielding up to 1.6x throughput improvement. Compared to state-of-the-art lossless compression approaches NeuZip and DFloat11, our design further improves throughput by up to 11x.

cs.AR↗

Otters++: A Time-to-first-spike Based Energy Efficient Optical Spiking Transformer

Spiking neural networks (SNNs) are promising for energy-efficient inference, and time-to-first-spike (TTFS) coding is especially attractive because each neuron fires at most once. In practice, however, this benefit is often reduced by the cost of computing a temporal decay term and multiplying it by the synaptic weight. We address this issue by turning a physical hardware "bug," the natural signal decay in optoelectronic devices, into the main computation of TTFS, named Otters++. Specifically, we use the measured decay of a custom In$_2$O$_3$ optoelectronic synapse to directly realize the TTFS temporal term, removing the need for explicit digital decay computation. To scale this idea to Transformer models, we establish a layer-wise functional equivalence between the Otters++ and a quantized neural network (QNN), and develop a hybrid training method that uses device-faithful SNN computation in the forward pass and QNN straight-through gradients through the equivalent QNN path in the backward pass, together with model distillation. This avoids differentiation through discrete first-spike events and reduces the over-sparsity problem in direct TTFS-SNN training. We further make training aware of measured device noise by sampling run-to-run variation, and refine the system-level energy model by accounting for device sharing and multi-hop communication. On GLUE dataset, Otters++ improves the average score to 84.17\% while maintaining a clear energy advantage over prior spiking Transformer baselines. These results show that physically grounded TTFS computing can be efficient, trainable, and robust under realistic hardware effects.

cs.AI↗

SePO: Self-Evolving Prompt Agent for System Prompt Optimization

System prompt optimization improves agent behavior without modifying the underlying model, yielding human-readable, model-agnostic instructions. Existing methods build a prompt agent that refines task agents' system prompts, yet leave the prompt agent's own system prompt hand-engineered and fixed. We propose Self-Evolving Prompt Optimization (SePO), which treats the prompt agent's own system prompt as an optimization target alongside task agents' system prompts. SePO adopts a self-referential design. A single prompt agent improves both task agents' system prompts and its own under an open-ended evolutionary search that maintains an archive of candidate prompts as stepping stones. Training proceeds in two stages: pre-training evolves the prompt agent on a multi-task pool, and fine-tuning then applies it to a target task. Across five benchmarks spanning math (AIME'25), abstract reasoning (ARC-AGI-1), graduate-level science (GPQA), code generation (MBPP), and logic puzzles (Sudoku), SePO consistently outperforms Manual-CoT, TextGrad, and MetaSPO, improving the average accuracy by 4.49 points compared to Manual-CoT. The prompt optimization skill from pre-training also generalizes to tasks beyond the pre-training mixture, rather than memorizing per-task prompts.

cs.CL↗

TileLoom: Automatic Dataflow Planning for Tile-Based Languages on Spatial Dataflow Accelerators

Spatial dataflow accelerators are a promising direction for next-generation computer systems because they can reduce the memory bottlenecks of traditional von Neumann machines such as CPUs and GPUs. They organize computation around explicit, compiler-managed data movement over on-chip networks, allowing operands to be forwarded directly between processing elements and reducing reliance on high-latency, bandwidth-limited global shared memory. However, their performance depends strongly on how workloads are mapped to hardware. Naive mappings can perform poorly, and most users rely on hand-tuned vendor libraries. Thus, despite their potential for high performance, energy efficiency, and cost efficiency, limited programmability remains a major barrier to wider adoption. This paper presents TileLoom, an MLIR-based end-to-end framework that compiles tile-based programs, such as Triton kernels, onto spatial dataflow architectures. Unlike compiler frameworks that focus on optimizing code generation within a single tile, TileLoom distributes tile instances across spatially distributed cores and exploits the on-chip network and distributed memories to increase data reuse and reduce communication. TileLoom introduces a hardware representation that captures interconnect topology, memory hierarchy, and compute capabilities, enabling both architecture-specific optimizations and support for diverse spatial dataflow targets. In experiments on two generations of Tenstorrent systems, TileLoom achieves performance comparable to vendor libraries on various kernels.

cs.DC↗

XtraMAC: An Efficient MAC Architecture for Mixed-Precision LLM Inference on FPGA

The widespread adoption of mixed-precision quantization in large language models (LLMs) has created demand for hardware that can efficiently perform multiply-accumulate (MAC) operations across mixed datatypes and switch datatypes at runtime. Existing FPGA-based MAC solutions fall short due to limitations in fixed-datatype design, inefficient spatial or temporal resource sharing, and poor support for mixed-precision execution. These limitations collectively lead to under-utilization of DSP resources, limiting achievable parallelism and throughput. In this work, we present XtraMAC, a novel MAC architecture that unifies integer, floating-point, and mixed-precision operations within a single, datatype-adaptive microarchitecture. XtraMAC decomposes all supported MAC formats into a shared integer mantissa product with lightweight sign and exponent handling, enabling dynamic operand packing and efficient DSP resource sharing with constant latency and initiation interval of one across all datatypes. Evaluated on an AMD Xilinx U55c FPGA, XtraMAC achieves 1.4-2.0x higher compute density, reduces per-operation LUT, FF, and DSP consumption by 27-51%, and delivers up to 1.9x greater energy efficiency and 1.2x speedup on representative mixed-precision LLM workloads. The implementation of XtraMAC is open-sourced at https://github.com/Xtra-Computing/XtraMAC.

cs.AR↗

ShiftLIF: Efficient Multi-Level Spiking Neurons with Power-of-Two Quantization

Spiking neural networks (SNNs) are promising for edge sensing due to their event-driven computation and temporal filtering capability. However, standard leaky integrate-and-fire (LIF) neurons communicate only through binary spikes, which severely limit representational capacity. Existing multi-level spiking neurons improve information transmission, but often rely on uniform quantization that mismatches membrane-potential distributions or introduces costly synaptic multiplications. In this paper, we propose ShiftLIF, a multi-level spiking neuron that maps membrane potentials to a logarithmically spaced power-of-two spike set. This design provides finer representation in the small-amplitude regime, where membrane potentials are densely concentrated, while enabling multiplier-free synaptic computation through bit-shift and accumulation operations. As a result, ShiftLIF improves spike-level expressiveness without sacrificing the hardware-friendly nature of standard SNN computation. We evaluate ShiftLIF on 10 datasets spanning wireless, acoustic, motion, and visual sensing tasks. Results show that ShiftLIF consistently matches or exceeds the accuracy of existing multi-level spiking neurons while maintaining synaptic energy consumption close to standard binary LIF. These results indicate that ShiftLIF provides a favorable accuracy-efficiency trade-off for cross-modal edge sensing.

cs.NE↗

SparrowSNN: A Hardware/software Co-design for Energy Efficient ECG Classification

Deep learning has driven significant technological advancements, but its high energy consumption limits its use on battery-operated edge devices. Spiking Neural Networks (SNNs) offer promising reductions in inference-time energy consumption. However, existing neuromorphic architectures optimize scalable, many-core NoC execution, suited to large models but mismatched to edge devices, and their prevalent integrate-and-fire neurons re-read weights across \(T\) timesteps, inflating data-movement and dynamic-control energy. To address this challenge, we propose SparrowSNN, an optimized end-to-end design tailored for edge applications. SparrowSNN proposes: (1) a hardware-friendly spike activation function SSF (Sum-Spike-and-Fire); (2) a customizable $μ$W-level-power quantized hybrid ANN-SNN model that can be designed per application; (3) a compact and low-power reconfigurable ASIC architecture, supporting the aforementioned designs. Evaluated on biomedical MIT-BIH ECG and DEAP EEG datasets, SparrowSNN achieves state-of-the-art accuracy with $20\times$ to $100\times$ lower energy consumption, significantly outperforming existing ultra-low power solutions.

cs.AR↗