Every FLOP Counts: Scaling a 300B Mixture-of-Experts LING LLM without Premium GPUs

In this technical report, we tackle the challenges of training large-scale Mixture of Experts (MoE) models, focusing on overcoming cost inefficiency and resource limitations prevalent in such systems. To address these issues, we present two differently sized MoE large language models (LLMs), namely Ling-Lite and Ling-Plus (referred to as "Bailing" in Chinese, spelled B\v{a}il\'ing in Pinyin). Ling-Lite contains 16.8 billion parameters with 2.75 billion activated parameters, while Ling-Plus boasts 290 billion parameters with 28.8 billion activated parameters. Both models exhibit comparable performance to leading industry benchmarks. This report offers actionable insights to improve the efficiency and accessibility of AI development in resource-constrained settings, promoting more scalable and sustainable technologies. Specifically, to reduce training costs for large-scale MoE models, we propose innovative methods for (1) optimization of model architecture and training processes, (2) refinement of training anomaly handling, and (3) enhancement of model evaluation efficiency. Additionally, leveraging high-quality data generated from knowledge graphs, our models demonstrate superior capabilities in tool use compared to other models. Ultimately, our experimental findings demonstrate that a 300B MoE LLM can be effectively trained on lower-performance devices while achieving comparable performance to models of a similar scale, including dense and MoE models. Compared to high-performance devices, utilizing a lower-specification hardware system during the pre-training phase demonstrates significant cost savings, reducing computing costs by approximately 20%. The models can be accessed at https://huggingface.co/inclusionAI.

MIDAS: Modeling Ground-Truth Distributions with Dark Knowledge for Domain Generalized Stereo Matching

Despite the significant advances in domain generalized stereo matching, existing methods still exhibit domain-specific preferences when transferring from synthetic to real domains, hindering their practical applications in complex and diverse scenarios. The probability distributions predicted by the stereo network naturally encode rich similarity and uncertainty information. Inspired by this observation, we propose to extract these two types of dark knowledge from the pre-trained network to model intuitive multi-modal ground-truth distributions for both edge and non-edge regions. To mitigate the inherent domain preferences of a single network, we adopt network ensemble and further distinguish between objective and biased knowledge in the Laplace parameter space. Finally, the objective knowledge and the original disparity labels are jointly modeled as a mixture of Laplacians to provide fine-grained supervision for the stereo network training. Extensive experiments demonstrate that: 1) Our method is generic and effectively improves the generalization of existing networks. 2) PCWNet with our method achieves the state-of-the-art generalization performance on both KITTI 2015 and 2012 datasets. 3) Our method outperforms existing methods in comprehensive ranking across four popular real-world datasets.

Large Memory Network for Recommendation

Modeling user behavior sequences in recommender systems is essential for understanding user preferences over time, enabling personalized and accurate recommendations for improving user retention and enhancing business values. Despite its significance, there are two challenges for current sequential modeling approaches. From the spatial dimension, it is difficult to mutually perceive similar users' interests for a generalized intention understanding; from the temporal dimension, current methods are generally prone to forgetting long-term interests due to the fixed-length input sequence. In this paper, we present Large Memory Network (LMN), providing a novel idea by compressing and storing user history behavior information in a large-scale memory block. With the elaborated online deployment strategy, the memory block can be easily scaled up to million-scale in the industry. Extensive offline comparison experiments, memory scaling up experiments, and online A/B test on Douyin E-Commerce Search (ECS) are performed, validating the superior performance of LMN. Currently, LMN has been fully deployed in Douyin ECS, serving millions of users each day.

SCKD: Semi-Supervised Cross-Modality Knowledge Distillation for 4D Radar Object Detection

3D object detection is one of the fundamental perception tasks for autonomous vehicles. Fulfilling such a task with a 4D millimeter-wave radar is very attractive since the sensor is able to acquire 3D point clouds similar to Lidar while maintaining robust measurements under adverse weather. However, due to the high sparsity and noise associated with the radar point clouds, the performance of the existing methods is still much lower than expected. In this paper, we propose a novel Semi-supervised Cross-modality Knowledge Distillation (SCKD) method for 4D radar-based 3D object detection. It characterizes the capability of learning the feature from a Lidar-radar-fused teacher network with semi-supervised distillation. We first propose an adaptive fusion module in the teacher network to boost its performance. Then, two feature distillation modules are designed to facilitate the cross-modality knowledge transfer. Finally, a semi-supervised output distillation is proposed to increase the effectiveness and flexibility of the distillation framework. With the same network structure, our radar-only student trained by SCKD boosts the mAP by 10.38% over the baseline and outperforms the state-of-the-art works on the VoD dataset. The experiment on ZJUODset also shows 5.12% mAP improvements on the moderate difficulty level over the baseline when extra unlabeled data are available. Code is available at https://github.com/Ruoyu-Xu/SCKD.

Traditional Chinese Medicine Case Analysis System for High-Level Semantic Abstraction: Optimized with Prompt and RAG

This paper details a technical plan for building a clinical case database for Traditional Chinese Medicine (TCM) using web scraping. Leveraging multiple platforms, including 360doc, we gathered over 5,000 TCM clinical cases, performed data cleaning, and structured the dataset with crucial fields such as patient details, pathogenesis, syndromes, and annotations. Using the $Baidu\_ERNIE\_Speed\_128K$ API, we removed redundant information and generated the final answers through the $DeepSeekv2$ API, outputting results in standard JSON format. We optimized data recall with RAG and rerank techniques during retrieval and developed a hybrid matching scheme. By combining two-stage retrieval method with keyword matching via Jieba, we significantly enhanced the accuracy of model outputs.

Tree-of-Table: Unleashing the Power of LLMs for Enhanced Large-Scale Table Understanding

The ubiquity and value of tables as semi-structured data across various domains necessitate advanced methods for understanding their complexity and vast amounts of information. Despite the impressive capabilities of large language models (LLMs) in advancing the natural language understanding frontier, their application to large-scale tabular data presents significant challenges, specifically regarding table size and complex intricate relationships. Existing works have shown promise with small-scale tables but often flounder when tasked with the complex reasoning required by larger, interconnected tables found in real-world scenarios. To address this gap, we introduce "Tree-of-Table", a novel approach designed to enhance LLMs' reasoning capabilities over large and complex tables. Our method employs Table Condensation and Decomposition to distill and reorganize relevant data into a manageable format, followed by the construction of a hierarchical Table-Tree that facilitates tree-structured reasoning. Through a meticulous Table-Tree Execution process, we systematically unravel the tree-structured reasoning chain to derive the solutions. Experiments across diverse datasets, including WikiTQ, TableFact, FeTaQA, and BIRD, demonstrate that Tree-of-Table sets a new benchmark with superior performance, showcasing remarkable efficiency and generalization capabilities in large-scale table reasoning.

Vision Language Models are In-Context Value Learners

Predicting temporal progress from visual trajectories is important for intelligent robots that can learn, adapt, and improve. However, learning such progress estimator, or temporal value function, across different tasks and domains requires both a large amount of diverse data and methods which can scale and generalize. To address these challenges, we present Generative Value Learning (\GVL), a universal value function estimator that leverages the world knowledge embedded in vision-language models (VLMs) to predict task progress. Naively asking a VLM to predict values for a video sequence performs poorly due to the strong temporal correlation between successive frames. Instead, GVL poses value estimation as a temporal ordering problem over shuffled video frames; this seemingly more challenging task encourages VLMs to more fully exploit their underlying semantic and temporal grounding capabilities to differentiate frames based on their perceived task progress, consequently producing significantly better value predictions. Without any robot or task specific training, GVL can in-context zero-shot and few-shot predict effective values for more than 300 distinct real-world tasks across diverse robot platforms, including challenging bimanual manipulation tasks. Furthermore, we demonstrate that GVL permits flexible multi-modal in-context learning via examples from heterogeneous tasks and embodiments, such as human videos. The generality of GVL enables various downstream applications pertinent to visuomotor policy learning, including dataset filtering, success detection, and advantage-weighted regression -- all without any model training or finetuning.

Frontiers in Intelligent Colonoscopy

Colonoscopy is currently one of the most sensitive screening methods for colorectal cancer. This study investigates the frontiers of intelligent colonoscopy techniques and their prospective implications for multimodal medical applications. With this goal, we begin by assessing the current data-centric and model-centric landscapes through four tasks for colonoscopic scene perception, including classification, detection, segmentation, and vision-language understanding. This assessment enables us to identify domain-specific challenges and reveals that multimodal research in colonoscopy remains open for further exploration. To embrace the coming multimodal era, we establish three foundational initiatives: a large-scale multimodal instruction tuning dataset ColonINST, a colonoscopy-designed multimodal language model ColonGPT, and a multimodal benchmark. To facilitate ongoing monitoring of this rapidly evolving field, we provide a public website for the latest updates: https://github.com/ai4colonoscopy/IntelliScope.

DCP: Learning Accelerator Dataflow for Neural Network via Propagation

Deep neural network (DNN) hardware (HW) accelerators have achieved great success in improving DNNs' performance and efficiency. One key reason is dataflow in executing a DNN layer, including on-chip data partitioning, computation parallelism, and scheduling policy, which have large impacts on latency and energy consumption. Unlike prior works that required considerable efforts from HW engineers to design suitable dataflows for different DNNs, this work proposes an efficient data-centric approach, named Dataflow Code Propagation (DCP), to automatically find the optimal dataflow for DNN layers in seconds without human effort. It has several attractive benefits that prior arts do not have. (i) We translate the HW dataflow configuration into a code representation in a unified dataflow coding space, which can be optimized by backpropagating gradients given a DNN layer or network. (ii) DCP learns a neural predictor to efficiently update the dataflow codes towards the desired gradient directions to minimize various optimization objectives e.g., latency and energy. (iii) It can be easily generalized to unseen HW configurations in a zero-shot or few-shot learning manner. For example, without using additional training data, DCP surpasses the GAMMA method that performs a full search using thousands of samples. Extensive experiments on several representative models such as MobileNet, ResNet, and ViT show that DCP outperforms its counterparts in various settings.

Generative Semantic Communication for Text-to-Speech Synthesis

Semantic communication is a promising technology to improve communication efficiency by transmitting only the semantic information of the source data. However, traditional semantic communication methods primarily focus on data reconstruction tasks, which may not be efficient for emerging generative tasks such as text-to-speech (TTS) synthesis. To address this limitation, this paper develops a novel generative semantic communication framework for TTS synthesis, leveraging generative artificial intelligence technologies. Firstly, we utilize a pre-trained large speech model called WavLM and the residual vector quantization method to construct two semantic knowledge bases (KBs) at the transmitter and receiver, respectively. The KB at the transmitter enables effective semantic extraction, while the KB at the receiver facilitates lifelike speech synthesis. Then, we employ a transformer encoder and a diffusion model to achieve efficient semantic coding without introducing significant communication overhead. Finally, numerical results demonstrate that our framework achieves much higher fidelity for the generated speech than four baselines, in both cases with additive white Gaussian noise channel and Rayleigh fading channel.