Bin Jia

ByteDance Seed, Jiaze Chen, Tiantian Fan et al. · bytedance

We introduce Seed1.5-Thinking, capable of reasoning through thinking before responding, resulting in improved performance on a wide range of benchmarks. Seed1.5-Thinking achieves 86.7 on AIME 2024, 55.0 on Codeforces and 77.3 on GPQA, demonstrating excellent reasoning abilities in STEM and coding. Beyond reasoning tasks, the method demonstrates notable generalization across diverse domains. For instance, it surpasses DeepSeek R1 by 8% in win rate on non-reasoning tasks, indicating its broader applicability. Compared to other state-of-the-art reasoning models, Seed1.5-Thinking is a Mixture-of-Experts (MoE) model with a relatively small size, featuring 20B activated and 200B total parameters. As part of our effort to assess generalized reasoning, we develop two internal benchmarks, BeyondAIME and Codeforces, both of which will be publicly released to support future research. Model trial link: https://www.volcengine.com/experience/ark.

Qianli Ma, Yaowei Zheng, Zhelun Shi et al.

Recent advances in large language models (LLMs) have driven impressive progress in omni-modal understanding and generation. However, training omni-modal LLMs remains a significant challenge due to the heterogeneous model architectures required to process diverse modalities, necessitating sophisticated system design for efficient large-scale training. Existing frameworks typically entangle model definition with parallel logic, incurring limited scalability and substantial engineering overhead for end-to-end omni-modal training. We present VeOmni, a modular and efficient training framework to accelerate the development of omni-modal LLMs. VeOmni introduces model-centric distributed recipes that decouples communication from computation, enabling efficient 3D parallelism on omni-modal LLMs. VeOmni also features a flexible configuration interface supporting seamless integration of new modalities with minimal code change. Using VeOmni, a omni-modal mixture-of-experts (MoE) model with 30B parameters can be trained with over 2,800 tokens/sec/GPU throughput and scale to 160K context lengths via 3D parallelism on 128 GPUs, showcasing its superior efficiency and scalability for training large omni-modal LLMs.

Yuqi Zheng, Ruidong Yan, Bin Jia et al.

In autonomous driving, the hybrid strategy of deep reinforcement learning and cooperative adaptive cruise control (CACC) can fully utilize the advantages of the two algorithms and significantly improve the performance of car following. However, it is challenging for the traditional hybrid strategy based on fixed coefficients to adapt to mixed traffic flow scenarios, which may decrease the performance and even lead to accidents. To address the above problems, a hybrid car following strategy based on an adaptive Kalman Filter is proposed by regarding CACC and Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithms. Different from traditional hybrid strategy based on fixed coefficients, the Kalman gain H, using as an adaptive coefficient, is derived from multi-timestep predictions and Monte Carlo Tree Search. At the end of study, simulation results with 4157745 timesteps indicate that, compared with the TD3 and HCFS algorithms, the proposed algorithm in this study can substantially enhance the safety of car following in mixed traffic flow without compromising the comfort and efficiency.

Xuanlei Zhao, Shenggan Cheng, Guangyang Lu et al.

Large deep learning models have achieved impressive performance across a range of applications. However, their large memory requirements, including parameter memory and activation memory, have become a significant challenge for their practical serving. While existing methods mainly address parameter memory, the importance of activation memory has been overlooked. Especially for long input sequences, activation memory is expected to experience a significant exponential growth as the length of sequences increases. In this approach, we propose AutoChunk, an automatic and adaptive compiler system that efficiently reduces activation memory for long sequence inference by chunk strategies. The proposed system generates chunk plans by optimizing through multiple stages. In each stage, the chunk search pass explores all possible chunk candidates and the chunk selection pass identifies the optimal one. At runtime, AutoChunk employs code generation to automatically apply chunk strategies. The experiments demonstrate that AutoChunk can reduce over 80\% of activation memory while maintaining speed loss within 10%, extend max sequence length by 3.2x to 11.7x, and outperform state-of-the-art methods by a large margin.

Ruidong Yan, Rui Jiang, Bin Jia et al.

Deep deterministic policy gradient (DDPG)-based car-following strategy can break through the constraints of the differential equation model due to the ability of exploration on complex environments. However, the car-following performance of DDPG is usually degraded by unreasonable reward function design, insufficient training, and low sampling efficiency. In order to solve this kind of problem, a hybrid car-following strategy based on DDPG and cooperative adaptive cruise control (CACC) is proposed. First, the car-following process is modeled as the Markov decision process to calculate CACC and DDPG simultaneously at each frame. Given a current state, two actions are obtained from CACC and DDPG, respectively. Then, an optimal action, corresponding to the one offering a larger reward, is chosen as the output of the hybrid strategy. Meanwhile, a rule is designed to ensure that the change rate of acceleration is smaller than the desired value. Therefore, the proposed strategy not only guarantees the basic performance of car-following through CACC but also makes full use of the advantages of exploration on complex environments via DDPG. Finally, simulation results show that the car-following performance of the proposed strategy is improved compared with that of DDPG and CACC.