Improved Sample Complexity for Global Convergence of Actor-Critic Algorithms

In this paper, we establish the global convergence of the actor-critic algorithm with a significantly improved sample complexity of $O(\epsilon^{-3})$, advancing beyond the existing local convergence results. Previous works provide local convergence guarantees with a sample complexity of $O(\epsilon^{-2})$ for bounding the squared gradient of the return, which translates to a global sample complexity of $O(\epsilon^{-4})$ using the gradient domination lemma. In contrast to traditional methods that employ decreasing step sizes for both the actor and critic, we demonstrate that a constant step size for the critic is sufficient to ensure convergence in expectation. This key insight reveals that using a decreasing step size for the actor alone is sufficient to handle the noise for both the actor and critic. Our findings provide theoretical support for the practical success of many algorithms that rely on constant step sizes.

EXAQ: Exponent Aware Quantization For LLMs Acceleration

Quantization has established itself as the primary approach for decreasing the computational and storage expenses associated with Large Language Models (LLMs) inference. The majority of current research emphasizes quantizing weights and activations to enable low-bit general-matrix-multiply (GEMM) operations, with the remaining non-linear operations executed at higher precision. In our study, we discovered that following the application of these techniques, the primary bottleneck in LLMs inference lies in the softmax layer. The softmax operation comprises three phases: exponent calculation, accumulation, and normalization, Our work focuses on optimizing the first two phases. We propose an analytical approach to determine the optimal clipping value for the input to the softmax function, enabling sub-4-bit quantization for LLMs inference. This method accelerates the calculations of both $e^x$ and $\sum(e^x)$ with minimal to no accuracy degradation. For example, in LLaMA1-30B, we achieve baseline performance with 2-bit quantization on the well-known "Physical Interaction: Question Answering" (PIQA) dataset evaluation. This ultra-low bit quantization allows, for the first time, an acceleration of approximately 4x in the accumulation phase. The combination of accelerating both $e^x$ and $\sum(e^x)$ results in a 36.9% acceleration in the softmax operation.

DropCompute: simple and more robust distributed synchronous training via compute variance reduction

Background: Distributed training is essential for large scale training of deep neural networks (DNNs). The dominant methods for large scale DNN training are synchronous (e.g. All-Reduce), but these require waiting for all workers in each step. Thus, these methods are limited by the delays caused by straggling workers. Results: We study a typical scenario in which workers are straggling due to variability in compute time. We find an analytical relation between compute time properties and scalability limitations, caused by such straggling workers. With these findings, we propose a simple yet effective decentralized method to reduce the variation among workers and thus improve the robustness of synchronous training. This method can be integrated with the widely used All-Reduce. Our findings are validated on large-scale training tasks using 200 Gaudi Accelerators.

High Probability Bounds for a Class of Nonconvex Algorithms with AdaGrad Stepsize

In this paper, we propose a new, simplified high probability analysis of AdaGrad for smooth, non-convex problems. More specifically, we focus on a particular accelerated gradient (AGD) template (Lan, 2020), through which we recover the original AdaGrad and its variant with averaging, and prove a convergence rate of $\mathcal O (1/ \sqrt{T})$ with high probability without the knowledge of smoothness and variance. We use a particular version of Freedman's concentration bound for martingale difference sequences (Kakade & Tewari, 2008) which enables us to achieve the best-known dependence of $\log (1 / \delta )$ on the probability margin $\delta$. We present our analysis in a modular way and obtain a complementary $\mathcal O (1 / T)$ convergence rate in the deterministic setting. To the best of our knowledge, this is the first high probability result for AdaGrad with a truly adaptive scheme, i.e., completely oblivious to the knowledge of smoothness and uniform variance bound, which simultaneously has best-known dependence of $\log( 1/ \delta)$. We further prove noise adaptation property of AdaGrad under additional noise assumptions.