Xiaoxiang Wang

DeepSeek-AI, Daya Guo, Dejian Yang et al. · stanford, tsinghua

We introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1. DeepSeek-R1-Zero, a model trained via large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step, demonstrates remarkable reasoning capabilities. Through RL, DeepSeek-R1-Zero naturally emerges with numerous powerful and intriguing reasoning behaviors. However, it encounters challenges such as poor readability, and language mixing. To address these issues and further enhance reasoning performance, we introduce DeepSeek-R1, which incorporates multi-stage training and cold-start data before RL. DeepSeek-R1 achieves performance comparable to OpenAI-o1-1217 on reasoning tasks. To support the research community, we open-source DeepSeek-R1-Zero, DeepSeek-R1, and six dense models (1.5B, 7B, 8B, 14B, 32B, 70B) distilled from DeepSeek-R1 based on Qwen and Llama.

DeepSeek-AI, Aixin Liu, Bei Feng et al. · pku

We present DeepSeek-V2, a strong Mixture-of-Experts (MoE) language model characterized by economical training and efficient inference. It comprises 236B total parameters, of which 21B are activated for each token, and supports a context length of 128K tokens. DeepSeek-V2 adopts innovative architectures including Multi-head Latent Attention (MLA) and DeepSeekMoE. MLA guarantees efficient inference through significantly compressing the Key-Value (KV) cache into a latent vector, while DeepSeekMoE enables training strong models at an economical cost through sparse computation. Compared with DeepSeek 67B, DeepSeek-V2 achieves significantly stronger performance, and meanwhile saves 42.5% of training costs, reduces the KV cache by 93.3%, and boosts the maximum generation throughput to 5.76 times. We pretrain DeepSeek-V2 on a high-quality and multi-source corpus consisting of 8.1T tokens, and further perform Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) to fully unlock its potential. Evaluation results show that, even with only 21B activated parameters, DeepSeek-V2 and its chat versions still achieve top-tier performance among open-source models.

DeepSeek-AI, Aixin Liu, Aoxue Mei et al.

We introduce DeepSeek-V3.2, a model that harmonizes high computational efficiency with superior reasoning and agent performance. The key technical breakthroughs of DeepSeek-V3.2 are as follows: (1) DeepSeek Sparse Attention (DSA): We introduce DSA, an efficient attention mechanism that substantially reduces computational complexity while preserving model performance in long-context scenarios. (2) Scalable Reinforcement Learning Framework: By implementing a robust reinforcement learning protocol and scaling post-training compute, DeepSeek-V3.2 performs comparably to GPT-5. Notably, our high-compute variant, DeepSeek-V3.2-Speciale, surpasses GPT-5 and exhibits reasoning proficiency on par with Gemini-3.0-Pro, achieving gold-medal performance in both the 2025 International Mathematical Olympiad (IMO) and the International Olympiad in Informatics (IOI). (3) Large-Scale Agentic Task Synthesis Pipeline: To integrate reasoning into tool-use scenarios, we developed a novel synthesis pipeline that systematically generates training data at scale. This methodology facilitates scalable agentic post-training, yielding substantial improvements in generalization and instruction-following robustness within complex, interactive environments.

DeepSeek-AI, Aixin Liu, Bei Feng et al. · stanford, tsinghua

We present DeepSeek-V3, a strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token. To achieve efficient inference and cost-effective training, DeepSeek-V3 adopts Multi-head Latent Attention (MLA) and DeepSeekMoE architectures, which were thoroughly validated in DeepSeek-V2. Furthermore, DeepSeek-V3 pioneers an auxiliary-loss-free strategy for load balancing and sets a multi-token prediction training objective for stronger performance. We pre-train DeepSeek-V3 on 14.8 trillion diverse and high-quality tokens, followed by Supervised Fine-Tuning and Reinforcement Learning stages to fully harness its capabilities. Comprehensive evaluations reveal that DeepSeek-V3 outperforms other open-source models and achieves performance comparable to leading closed-source models. Despite its excellent performance, DeepSeek-V3 requires only 2.788M H800 GPU hours for its full training. In addition, its training process is remarkably stable. Throughout the entire training process, we did not experience any irrecoverable loss spikes or perform any rollbacks. The model checkpoints are available at https://github.com/deepseek-ai/DeepSeek-V3.

Zhaoxing Zhang, Junda Cheng, Gangwei Xu et al.

Recent approaches to VO have significantly improved performance by using deep networks to predict optical flow between video frames. However, existing methods still suffer from noisy and inconsistent flow matching, making it difficult to handle challenging scenarios and long-sequence estimation. To overcome these challenges, we introduce Spatio-Temporal Visual Odometry (STVO), a novel deep network architecture that effectively leverages inherent spatio-temporal cues to enhance the accuracy and consistency of multi-frame flow matching. With more accurate and consistent flow matching, STVO can achieve better pose estimation through the bundle adjustment (BA). Specifically, STVO introduces two innovative components: 1) the Temporal Propagation Module that utilizes multi-frame information to extract and propagate temporal cues across adjacent frames, maintaining temporal consistency; 2) the Spatial Activation Module that utilizes geometric priors from the depth maps to enhance spatial consistency while filtering out excessive noise and incorrect matches. Our STVO achieves state-of-the-art performance on TUM-RGBD, EuRoc MAV, ETH3D and KITTI Odometry benchmarks. Notably, it improves accuracy by 77.8% on ETH3D benchmark and 38.9% on KITTI Odometry benchmark over the previous best methods.

Xiaoxiang Wang, Jiaxin Liu, Miaojie Feng et al.

3D Multi-Object Tracking (MOT), a fundamental component of environmental perception, is essential for intelligent systems like autonomous driving and robotic sensing. Although Tracking-by-Detection frameworks have demonstrated excellent performance in recent years, their application in real-world scenarios faces significant challenges. Object movement in complex environments is often highly nonlinear, while existing methods typically rely on linear approximations of motion. Furthermore, system noise is frequently modeled as a Gaussian distribution, which fails to capture the true complexity of the noise dynamics. These oversimplified modeling assumptions can lead to significant reductions in tracking precision. To address this, we propose a GRU-based MOT method, which introduces a learnable Kalman filter into the motion module. This approach is able to learn object motion characteristics through data-driven learning, thereby avoiding the need for manual model design and model error. At the same time, to avoid abnormal supervision caused by the wrong association between annotations and trajectories, we design a semi-supervised learning strategy to accelerate the convergence speed and improve the robustness of the model. Evaluation experiment on the nuScenes and Argoverse2 datasets demonstrates that our system exhibits superior performance and significant potential compared to traditional TBD methods.

Liyue Chen, Xiaoxiang Wang, Leye Wang

Contextual features are important data sources for building citywide crowd mobility prediction models. However, the difficulty of applying context lies in the unknown generalizability of contextual features (e.g., weather, holiday, and points of interests) and context modeling techniques across different scenarios. In this paper, we present a unified analytic framework and a large-scale benchmark for evaluating context generalizability. The benchmark includes crowd mobility data, contextual data, and advanced prediction models. We conduct comprehensive experiments in several crowd mobility prediction tasks such as bike flow, metro passenger flow, and electric vehicle charging demand. Our results reveal several important observations: (1) Using more contextual features may not always result in better prediction with existing context modeling techniques; in particular, the combination of holiday and temporal position can provide more generalizable beneficial information than other contextual feature combinations. (2) In context modeling techniques, using a gated unit to incorporate raw contextual features into the deep prediction model has good generalizability. Besides, we offer several suggestions about incorporating contextual factors for building crowd mobility prediction applications. From our findings, we call for future research efforts devoted to developing new context modeling solutions.