arXiv subjects

Jiabin Yang

Publications and source records attributed to Jiabin Yang.

5 recordsLinked to original sources

arXiv2026

ERNIE 5.0 Technical Report

In this report, we introduce ERNIE 5.0, a natively autoregressive foundation model desinged for unified multimodal understanding and generation across text, image, video, and audio. All modalities are trained from scratch under a unified next-group-of-tokens prediction objective, based on an ultra-sparse mixture-of-experts (MoE) architecture with modality-agnostic expert routing. To address practical challenges in large-scale deployment under diverse resource constraints, ERNIE 5.0 adopts a novel elastic training paradigm. Within a single pre-training run, the model learns a family of sub-models with varying depths, expert capacities, and routing sparsity, enabling flexible trade-offs among performance, model size, and inference latency in memory- or time-constrained scenarios. Moreover, we systematically address the challenges of scaling reinforcement learning to unified foundation models, thereby guaranteeing efficient and stable post-training under ultra-sparse MoE architectures and diverse multimodal settings. Extensive experiments demonstrate that ERNIE 5.0 achieves strong and balanced performance across multiple modalities. To the best of our knowledge, among publicly disclosed models, ERNIE 5.0 represents the first production-scale realization of a trillion-parameter unified autoregressive model that supports both multimodal understanding and generation. To facilitate further research, we present detailed visualizations of modality-agnostic expert routing in the unified model, alongside comprehensive empirical analysis of elastic training, aiming to offer profound insights to the community.

cs.CL↗

arXiv2025

AdaGC: Enhancing LLM Pretraining Stability via Adaptive Gradient Clipping

Loss spikes remain a persistent obstacle in large-scale language model pretraining. While previous research has attempted to identify the root cause of loss spikes by investigating individual factors, we observe that, in practice, such spikes are typically triggered by the confluence of heterogeneous factors. Empirically, loss spikes may arise from a combination of data outliers, hardware or transient computational faults, numerical precision issues, and hyperparameter settings. Regardless of the underlying cause, these spikes manifest as unstable optimizer updates, as abnormal gradients contaminate both first- and second-moment states. In this paper, we propose a principled gradient-centric remedy: AdaGC, an adaptive per-tensor gradient clipping scheme that mitigates such contamination by bounding gradient norms relative to a tensor-wise exponential moving average of their historical clipped values. AdaGC is optimizer-agnostic, introduces negligible memory overhead, and reduces communication costs compared to GlobalGC, particularly in hybrid-parallel distributed training. Experiments on Llama-2 7B, Mixtral 8x1B, and ERNIE 10B-A1.4B demonstrate that AdaGC robustly eliminates training instabilities, consistently reducing spike scores to zero for all models and improving downstream accuracy over GlobalGC by 1.32%, 1.27%, and 2.48%, respectively. Furthermore, AdaGC seamlessly integrates with optimizers such as Muon and Lion, consistently yielding higher average accuracy and zero spike scores. The code is available at https://github.com/PaddlePaddle/PaddleFleet (see Research/AdaGC).

cs.LG↗

arXiv2024

FlashMask: Efficient and Rich Mask Extension of FlashAttention

The computational and memory demands of vanilla attention scale quadratically with the sequence length $N$, posing significant challenges for processing long sequences in Transformer models. FlashAttention alleviates these challenges by eliminating the $O(N^2)$ memory dependency and reducing attention latency through IO-aware memory optimizations. However, its native support for certain attention mask types is limited, and it does not inherently accommodate more complex masking requirements. Previous approaches resort to using dense masks with $O(N^2)$ memory complexity, leading to inefficiencies. In this paper, we propose FlashMask, an extension of FlashAttention that introduces a column-wise sparse representation of attention masks. This approach efficiently represents a wide range of mask types and facilitates the development of optimized kernel implementations. By adopting this novel representation, FlashMask achieves linear memory complexity $O(N)$, suitable for modeling long-context sequences. Moreover, this representation enables kernel optimizations that eliminate unnecessary computations by leveraging sparsity in the attention mask, without sacrificing computational accuracy, resulting in higher computational efficiency. We evaluate FlashMask's performance in fine-tuning and alignment training of LLMs such as SFT, LoRA, DPO, and RM. FlashMask achieves significant throughput improvements, with end-to-end speedups ranging from 1.65x to 3.22x compared to existing FlashAttention dense method. Additionally, our kernel-level comparisons demonstrate that FlashMask surpasses the latest counterpart, FlexAttention, by 12.1% to 60.7% in terms of kernel TFLOPs/s, achieving 37.8% to 62.3% of the theoretical maximum FLOPs/s on the A100 GPU. The code is open-sourced on PaddlePaddle and integrated into PaddleNLP, supporting models with over 100 billion parameters for contexts up to 128K tokens.

cs.LG↗

arXiv2023

DeePMD-kit v2: A software package for Deep Potential models

DeePMD-kit is a powerful open-source software package that facilitates molecular dynamics simulations using machine learning potentials (MLP) known as Deep Potential (DP) models. This package, which was released in 2017, has been widely used in the fields of physics, chemistry, biology, and material science for studying atomistic systems. The current version of DeePMD-kit offers numerous advanced features such as DeepPot-SE, attention-based and hybrid descriptors, the ability to fit tensile properties, type embedding, model deviation, Deep Potential - Range Correction (DPRc), Deep Potential Long Range (DPLR), GPU support for customized operators, model compression, non-von Neumann molecular dynamics (NVNMD), and improved usability, including documentation, compiled binary packages, graphical user interfaces (GUI), and application programming interfaces (API). This article presents an overview of the current major version of the DeePMD-kit package, highlighting its features and technical details. Additionally, the article benchmarks the accuracy and efficiency of different models and discusses ongoing developments.

physics.chem-ph↗

arXiv2019

A kilo-Ampere level HTS flux pump

This paper reports a newly developed high current transformer-rectifier High-Tc Superconducting (HTS) flux pump switched by dynamic resistance. A quasi-persistent current of over 1.1 kA has been achieved at 77 K using the device, which is the highest reported operating current by any HTS flux pumps to date. The size of the device is much smaller than traditional current leads and power supplies at the same current level. Parallel YBCO coated conductors are used in the transformer secondary winding as well as in the superconducting load coil to achieve high current. The output current is limited by the critical current of the load rather than the flux pump itself. Moreover, at over 1 kA current level, the device can maintain high flux injection accuracy, and the overall flux ripple is less than 0.2 mili-Weber. The work has shown the potential of using the device to operate high field HTS magnets in ultra-high quasi-persistent current mode, thus substantially reducing the inductance, size, weight, and cost of high field magnets, making them more accessible. It also indicates that the device is promising for powering HTS NMR/MRI magnets, in which the requirement for magnetic field satiability is demanding.

physics.app-ph↗