Large language models (LLMs) deliver impressive performance but require large amounts of energy. In this work, we present a MatMul-free LLM architecture adapted for Intel's neuromorphic processor, Loihi 2. Our approach leverages Loihi 2's support for low-precision, event-driven computation and stateful processing. Our hardware-aware quantized model on GPU demonstrates that a 370M parameter MatMul-free model can be quantized with no accuracy loss. Based on preliminary results, we report up to 3x higher throughput with 2x less energy, compared to transformer-based LLMs on an edge GPU, with significantly better scaling. Further hardware optimizations will increase throughput and decrease energy consumption. These results show the potential of neuromorphic hardware for efficient inference and pave the way for efficient reasoning models capable of generating complex, long-form text rapidly and cost-effectively.

With the increasing demand for practical applications of Large Language Models (LLMs), many attention-efficient models have been developed to balance performance and computational cost. However, the adversarial robustness of these models remains under-explored. In this work, we design a framework to investigate the trade-off between efficiency, performance, and adversarial robustness of LLMs and conduct extensive experiments on three prominent models with varying levels of complexity and efficiency -- Transformer++, Gated Linear Attention (GLA) Transformer, and MatMul-Free LM -- utilizing the GLUE and AdvGLUE datasets. The AdvGLUE dataset extends the GLUE dataset with adversarial samples designed to challenge model robustness. Our results show that while the GLA Transformer and MatMul-Free LM achieve slightly lower accuracy on GLUE tasks, they demonstrate higher efficiency and either superior or comparative robustness on AdvGLUE tasks compared to Transformer++ across different attack levels. These findings highlight the potential of simplified architectures to achieve a compelling balance between efficiency, performance, and adversarial robustness, offering valuable insights for applications where resource constraints and resilience to adversarial attacks are critical.

Matrix multiplication (MatMul) typically dominates the overall computational cost of large language models (LLMs). This cost only grows as LLMs scale to larger embedding dimensions and context lengths. In this work, we show that MatMul operations can be completely eliminated from LLMs while maintaining strong performance at billion-parameter scales. Our experiments show that our proposed MatMul-free models achieve performance on-par with state-of-the-art Transformers that require far more memory during inference at a scale up to at least 2.7B parameters. We investigate the scaling laws and find that the performance gap between our MatMul-free models and full precision Transformers narrows as the model size increases. We also provide a GPU-efficient implementation of this model which reduces memory usage by up to 61% over an unoptimized baseline during training. By utilizing an optimized kernel during inference, our model's memory consumption can be reduced by more than 10x compared to unoptimized models. To properly quantify the efficiency of our architecture, we build a custom hardware solution on an FPGA which exploits lightweight operations beyond what GPUs are capable of. We processed billion-parameter scale models at 13W beyond human readable throughput, moving LLMs closer to brain-like efficiency. This work not only shows how far LLMs can be stripped back while still performing effectively, but also points at the types of operations future accelerators should be optimized for in processing the next generation of lightweight LLMs. Our code implementation is available at https://github.com/ridgerchu/matmulfreellm.