FastCuRL: Curriculum Reinforcement Learning with Progressive Context Extension for Efficient Training R1-like Reasoning Models

In this paper, we propose \textbf{\textsc{FastCuRL}}, a simple yet efficient \textbf{Cu}rriculum \textbf{R}einforcement \textbf{L}earning approach with context window extending strategy to accelerate the reinforcement learning training efficiency for R1-like reasoning models while enhancing their performance in tackling complex reasoning tasks with long chain-of-thought rationales, particularly with a 1.5B parameter language model. \textbf{\textsc{FastCuRL}} consists of two main procedures: length-aware training data segmentation and context window extension training. Specifically, the former first splits the original training data into three different levels by the input prompt length, and then the latter leverages segmented training datasets with a progressively increasing context window length to train the reasoning model. Experimental results demonstrate that \textbf{\textsc{FastCuRL}}-1.5B-Preview surpasses DeepScaleR-1.5B-Preview across all five datasets (including MATH 500, AIME 2024, AMC 2023, Minerva Math, and OlympiadBench) while only utilizing 50\% of training steps. Furthermore, all training stages for FastCuRL-1.5B-Preview are completed using just a single node with 8 GPUs.

Towards Superior Quantization Accuracy: A Layer-sensitive Approach

Large Vision and Language Models have exhibited remarkable human-like intelligence in tasks such as natural language comprehension, problem-solving, logical reasoning, and knowledge retrieval. However, training and serving these models require substantial computational resources, posing a significant barrier to their widespread application and further research. To mitigate this challenge, various model compression techniques have been developed to reduce computational requirements. Nevertheless, existing methods often employ uniform quantization configurations, failing to account for the varying difficulties across different layers in quantizing large neural network models. This paper tackles this issue by leveraging layer-sensitivity features, such as activation sensitivity and weight distribution Kurtosis, to identify layers that are challenging to quantize accurately and allocate additional memory budget. The proposed methods, named SensiBoost and KurtBoost, respectively, demonstrate notable improvement in quantization accuracy, achieving up to 9% lower perplexity with only a 2% increase in memory budget on LLama models compared to the baseline.

Long-Term Interest Clock: Fine-Grained Time Perception in Streaming Recommendation System

User interests manifest a dynamic pattern within the course of a day, e.g., a user usually favors soft music at 8 a.m. but may turn to ambient music at 10 p.m. To model dynamic interests in a day, hour embedding is widely used in traditional daily-trained industrial recommendation systems. However, its discreteness can cause periodical online patterns and instability in recent streaming recommendation systems. Recently, Interest Clock has achieved remarkable performance in streaming recommendation systems. Nevertheless, it models users' dynamic interests in a coarse-grained manner, merely encoding users' discrete interests of 24 hours from short-term behaviors. In this paper, we propose a fine-grained method for perceiving time information for streaming recommendation systems, named Long-term Interest Clock (LIC). The key idea of LIC is adaptively calculating current user interests by taking into consideration the relevance of long-term behaviors around current time (e.g., 8 a.m.) given a candidate item. LIC consists of two modules: (1) Clock-GSU retrieves a sub-sequence by searching through long-term behaviors, using query information from a candidate item and current time, (2) Clock-ESU employs a time-gap-aware attention mechanism to aggregate sub-sequence with the candidate item. With Clock-GSU and Clock-ESU, LIC is capable of capturing users' dynamic fine-grained interests from long-term behaviors. We conduct online A/B tests, obtaining +0.122% improvements on user active days. Besides, the extended offline experiments show improvements as well. Long-term Interest Clock has been integrated into Douyin Music App's recommendation system.

AdaF^2M^2: Comprehensive Learning and Responsive Leveraging Features in Recommendation System

Feature modeling, which involves feature representation learning and leveraging, plays an essential role in industrial recommendation systems. However, the data distribution in real-world applications usually follows a highly skewed long-tail pattern due to the popularity bias, which easily leads to over-reliance on ID-based features, such as user/item IDs and ID sequences of interactions. Such over-reliance makes it hard for models to learn features comprehensively, especially for those non-ID meta features, e.g., user/item characteristics. Further, it limits the feature leveraging ability in models, getting less generalized and more susceptible to data noise. Previous studies on feature modeling focus on feature extraction and interaction, hardly noticing the problems brought about by the long-tail data distribution. To achieve better feature representation learning and leveraging on real-world data, we propose a model-agnostic framework AdaF^2M^2, short for Adaptive Feature Modeling with Feature Mask. The feature-mask mechanism helps comprehensive feature learning via multi-forward training with augmented samples, while the adapter applies adaptive weights on features responsive to different user/item states. By arming base models with AdaF^2M^2, we conduct online A/B tests on multiple recommendation scenarios, obtaining +1.37% and +1.89% cumulative improvements on user active days and app duration respectively. Besides, the extended offline experiments on different models show improvements as well. AdaF$^2$M$^2$ has been widely deployed on both retrieval and ranking tasks in multiple applications of Douyin Group, indicating its superior effectiveness and universality.

Real-time Indexing for Large-scale Recommendation by Streaming Vector Quantization Retriever

Retrievers, which form one of the most important recommendation stages, are responsible for efficiently selecting possible positive samples to the later stages under strict latency limitations. Because of this, large-scale systems always rely on approximate calculations and indexes to roughly shrink candidate scale, with a simple ranking model. Considering simple models lack the ability to produce precise predictions, most of the existing methods mainly focus on incorporating complicated ranking models. However, another fundamental problem of index effectiveness remains unresolved, which also bottlenecks complication. In this paper, we propose a novel index structure: streaming Vector Quantization model, as a new generation of retrieval paradigm. Streaming VQ attaches items with indexes in real time, granting it immediacy. Moreover, through meticulous verification of possible variants, it achieves additional benefits like index balancing and reparability, enabling it to support complicated ranking models as existing approaches. As a lightweight and implementation-friendly architecture, streaming VQ has been deployed and replaced all major retrievers in Douyin and Douyin Lite, resulting in remarkable user engagement gain.

Robust Low-Light Human Pose Estimation through Illumination-Texture Modulation

As critical visual details become obscured, the low visibility and high ISO noise in extremely low-light images pose a significant challenge to human pose estimation. Current methods fail to provide high-quality representations due to reliance on pixel-level enhancements that compromise semantics and the inability to effectively handle extreme low-light conditions for robust feature learning. In this work, we propose a frequency-based framework for low-light human pose estimation, rooted in the "divide-and-conquer" principle. Instead of uniformly enhancing the entire image, our method focuses on task-relevant information. By applying dynamic illumination correction to the low-frequency components and low-rank denoising to the high-frequency components, we effectively enhance both the semantic and texture information essential for accurate pose estimation. As a result, this targeted enhancement method results in robust, high-quality representations, significantly improving pose estimation performance. Extensive experiments demonstrating its superiority over state-of-the-art methods in various challenging low-light scenarios.

AgentPose: Progressive Distribution Alignment via Feature Agent for Human Pose Distillation

Pose distillation is widely adopted to reduce model size in human pose estimation. However, existing methods primarily emphasize the transfer of teacher knowledge while often neglecting the performance degradation resulted from the curse of capacity gap between teacher and student. To address this issue, we propose AgentPose, a novel pose distillation method that integrates a feature agent to model the distribution of teacher features and progressively aligns the distribution of student features with that of the teacher feature, effectively overcoming the capacity gap and enhancing the ability of knowledge transfer. Our comprehensive experiments conducted on the COCO dataset substantiate the effectiveness of our method in knowledge transfer, particularly in scenarios with a high capacity gap.

DuoCast: Duo-Probabilistic Meteorology-Aware Model for Extended Precipitation Nowcasting

Recently, extended short-term precipitation nowcasting struggles with decreasing precision because of insufficient consideration of meteorological knowledge, such as weather fronts which significantly influence precipitation intensity, duration, and spatial distribution. Therefore, in this paper, we present DuoCast, a novel dual-probabilistic meteorology-aware model designed to address both broad weather evolution and micro-scale fluctuations using two diffusion models, PrecipFlow and MicroDynamic, respectively. Our PrecipFlow model captures evolution trends through an Extreme Precipitation-Aware Encoder (EPA-Encoder), which includes AirConvolution and FrontAttention blocks to process two levels of precipitation data: general and extreme. The output conditions a UNet-based diffusion to produce prediction maps enriched with weather front information. The MicroDynamic model further refines the results to capture micro-scale variability. Extensive experiments on four public benchmarks demonstrate the effectiveness of our DuoCast, achieving superior performance over state-of-the-art methods. Our code is available at https://github.com/ph-w2000/DuoCast.

DaYu: Data-Driven Model for Geostationary Satellite Observed Cloud Images Forecasting

In the past few years, Artificial Intelligence (AI)-based weather forecasting methods have widely demonstrated strong competitiveness among the weather forecasting systems. However, these methods are insufficient for high-spatial-resolution short-term nowcasting within 6 hours, which is crucial for warning short-duration, mesoscale and small-scale weather events. Geostationary satellite remote sensing provides detailed, high spatio-temporal and all-day observations, which can address the above limitations of existing methods. Therefore, this paper proposed an advanced data-driven thermal infrared cloud images forecasting model, "DaYu." Unlike existing data-driven weather forecasting models, DaYu is specifically designed for geostationary satellite observations, with a temporal resolution of 0.5 hours and a spatial resolution of ${0.05}^\circ$ $\times$ ${0.05}^\circ$. DaYu is based on a large-scale transformer architecture, which enables it to capture fine-grained cloud structures and learn fast-changing spatio-temporal evolution features effectively. Moreover, its attention mechanism design achieves a balance in computational complexity, making it practical for applications. DaYu not only achieves accurate forecasts up to 3 hours with a correlation coefficient higher than 0.9, 6 hours higher than 0.8, and 12 hours higher than 0.7, but also detects short-duration, mesoscale, and small-scale weather events with enhanced detail, effectively addressing the shortcomings of existing methods in providing detailed short-term nowcasting within 6 hours. Furthermore, DaYu has significant potential in short-term climate disaster prevention and mitigation.

Hunyuan-Large: An Open-Source MoE Model with 52 Billion Activated Parameters by Tencent

In this paper, we introduce Hunyuan-Large, which is currently the largest open-source Transformer-based mixture of experts model, with a total of 389 billion parameters and 52 billion activation parameters, capable of handling up to 256K tokens. We conduct a thorough evaluation of Hunyuan-Large's superior performance across various benchmarks including language understanding and generation, logical reasoning, mathematical problem-solving, coding, long-context, and aggregated tasks, where it outperforms LLama3.1-70B and exhibits comparable performance when compared to the significantly larger LLama3.1-405B model. Key practice of Hunyuan-Large include large-scale synthetic data that is orders larger than in previous literature, a mixed expert routing strategy, a key-value cache compression technique, and an expert-specific learning rate strategy. Additionally, we also investigate the scaling laws and learning rate schedule of mixture of experts models, providing valuable insights and guidances for future model development and optimization. The code and checkpoints of Hunyuan-Large are released to facilitate future innovations and applications. Codes: https://github.com/Tencent/Hunyuan-Large Models: https://huggingface.co/tencent/Tencent-Hunyuan-Large