Jidong Zhai

Aohan Zeng, Xiao Liu, Zhengxiao Du et al. · tsinghua

We introduce GLM-130B, a bilingual (English and Chinese) pre-trained language model with 130 billion parameters. It is an attempt to open-source a 100B-scale model at least as good as GPT-3 (davinci) and unveil how models of such a scale can be successfully pre-trained. Over the course of this effort, we face numerous unexpected technical and engineering challenges, particularly on loss spikes and divergence. In this paper, we introduce the training process of GLM-130B including its design choices, training strategies for both efficiency and stability, and engineering efforts. The resultant GLM-130B model offers significant outperformance over GPT-3 175B (davinci) on a wide range of popular English benchmarks while the performance advantage is not observed in OPT-175B and BLOOM-176B. It also consistently and significantly outperforms ERNIE TITAN 3.0 260B -- the largest Chinese language model -- across related benchmarks. Finally, we leverage a unique scaling property of GLM-130B to reach INT4 quantization without post training, with almost no performance loss, making it the first among 100B-scale models and more importantly, allowing its effective inference on 4$\times$RTX 3090 (24G) or 8$\times$RTX 2080 Ti (11G) GPUs, the most affordable GPUs required for using 100B-scale models. The GLM-130B model weights are publicly accessible and its code, training logs, related toolkit, and lessons learned are open-sourced at \url{https://github.com/THUDM/GLM-130B/}.

Sha Yuan, Hanyu Zhao, Shuai Zhao et al. · bytedance, pku

With the rapid development of deep learning, training Big Models (BMs) for multiple downstream tasks becomes a popular paradigm. Researchers have achieved various outcomes in the construction of BMs and the BM application in many fields. At present, there is a lack of research work that sorts out the overall progress of BMs and guides the follow-up research. In this paper, we cover not only the BM technologies themselves but also the prerequisites for BM training and applications with BMs, dividing the BM review into four parts: Resource, Models, Key Technologies and Application. We introduce 16 specific BM-related topics in those four parts, they are Data, Knowledge, Computing System, Parallel Training System, Language Model, Vision Model, Multi-modal Model, Theory&Interpretability, Commonsense Reasoning, Reliability&Security, Governance, Evaluation, Machine Translation, Text Generation, Dialogue and Protein Research. In each topic, we summarize clearly the current studies and propose some future research directions. At the end of this paper, we conclude the further development of BMs in a more general view.

Lunyiu Nie, Shulin Cao, Jiaxin Shi et al. · tsinghua

Subject to the huge semantic gap between natural and formal languages, neural semantic parsing is typically bottlenecked by its complexity of dealing with both input semantics and output syntax. Recent works have proposed several forms of supplementary supervision but none is generalized across multiple formal languages. This paper proposes a unified intermediate representation (IR) for graph query languages, named GraphQ IR. It has a natural-language-like expression that bridges the semantic gap and formally defined syntax that maintains the graph structure. Therefore, a neural semantic parser can more precisely convert user queries into GraphQ IR, which can be later losslessly compiled into various downstream graph query languages. Extensive experiments on several benchmarks including KQA Pro, Overnight, GrailQA, and MetaQA-Cypher under standard i.i.d., out-of-distribution, and low-resource settings validate GraphQ IR's superiority over the previous state-of-the-arts with a maximum 11% accuracy improvement.

Kezhao Huang, Haitian Jiang, Minjie Wang et al.

A key performance bottleneck when training graph neural network (GNN) models on large, real-world graphs is loading node features onto a GPU. Due to limited GPU memory, expensive data movement is necessary to facilitate the storage of these features on alternative devices with slower access (e.g. CPU memory). Moreover, the irregularity of graph structures contributes to poor data locality which further exacerbates the problem. Consequently, existing frameworks capable of efficiently training large GNN models usually incur a significant accuracy degradation because of the currently-available shortcuts involved. To address these limitations, we instead propose FreshGNN, a general-purpose GNN mini-batch training framework that leverages a historical cache for storing and reusing GNN node embeddings instead of re-computing them through fetching raw features at every iteration. Critical to its success, the corresponding cache policy is designed, using a combination of gradient-based and staleness criteria, to selectively screen those embeddings which are relatively stable and can be cached, from those that need to be re-computed to reduce estimation errors and subsequent downstream accuracy loss. When paired with complementary system enhancements to support this selective historical cache, FreshGNN is able to accelerate the training speed on large graph datasets such as ogbn-papers100M and MAG240M by 3.4x up to 20.5x and reduce the memory access by 59%, with less than 1% influence on test accuracy.

Lunyiu Nie, Jiuding Sun, Yanlin Wang et al.

The recent prevalence of pretrained language models (PLMs) has dramatically shifted the paradigm of semantic parsing, where the mapping from natural language utterances to structured logical forms is now formulated as a Seq2Seq task. Despite the promising performance, previous PLM-based approaches often suffer from hallucination problems due to their negligence of the structural information contained in the sentence, which essentially constitutes the key semantics of the logical forms. Furthermore, most works treat PLM as a black box in which the generation process of the target logical form is hidden beneath the decoder modules, which greatly hinders the model's intrinsic interpretability. To address these two issues, we propose to incorporate the current PLMs with a hierarchical decoder network. By taking the first-principle structures as the semantic anchors, we propose two novel intermediate supervision tasks, namely Semantic Anchor Extraction and Semantic Anchor Alignment, for training the hierarchical decoders and probing the model intermediate representations in a self-adaptive manner alongside the fine-tuning process. We conduct intensive experiments on several semantic parsing benchmarks and demonstrate that our approach can consistently outperform the baselines. More importantly, by analyzing the intermediate representations of the hierarchical decoders, our approach also makes a huge step toward the intrinsic interpretability of PLMs in the domain of semantic parsing.

Liyan Zheng, Haojie Wang, Jidong Zhai et al.

Boosting the runtime performance of deep neural networks (DNNs) is critical due to their wide adoption in real-world tasks. Existing approaches to optimizing the tensor algebra expression of a DNN only consider expressions representable by a fixed set of predefined operators, missing possible optimization opportunities between general expressions. We propose OLLIE, the first derivation-based tensor program optimizer. OLLIE optimizes tensor programs by leveraging transformations between general tensor algebra expressions, enabling a significantly larger expression search space that includes those supported by prior work as special cases. OLLIE uses a hybrid derivation-based optimizer that effectively combines explorative and guided derivations to quickly discover highly optimized expressions. Evaluation on seven DNNs shows that OLLIE can outperform existing optimizers by up to 2.73$\times$ (1.46$\times$ on average) on an A100 GPU and up to 2.68$\times$ (1.51$\times$) on a V100 GPU, respectively.

Zixuan Ma, Haojie Wang, Jingze Xing et al.

Deep neural networks (DNNs) are of critical use in different domains. To accelerate DNN computation, tensor compilers are proposed to generate efficient code on different domain-specific accelerators. Existing tensor compilers mainly focus on optimizing computation efficiency. However, memory access is becoming a key performance bottleneck because the computational performance of accelerators is increasing much faster than memory performance. The lack of direct description of memory access and data dependence in current tensor compilers' intermediate representation (IR) brings significant challenges to generate memory-efficient code. In this paper, we propose IntelliGen, a tensor compiler that can generate high-performance code for memory-intensive operators by considering both computation and data movement optimizations. IntelliGen represent a DNN program using GIR, which includes primitives indicating its computation, data movement, and parallel strategies. This information will be further composed as an instruction-level dataflow graph to perform holistic optimizations by searching different memory access patterns and computation operations, and generating memory-efficient code on different hardware. We evaluate IntelliGen on NVIDIA GPU, AMD GPU, and Cambricon MLU, showing speedup up to 1.97x, 2.93x, and 16.91x(1.28x, 1.23x, and 2.31x on average), respectively, compared to current most performant frameworks.

GLM-4. 5 Team, Aohan Zeng, Xin Lv et al.

We present GLM-4.5, an open-source Mixture-of-Experts (MoE) large language model with 355B total parameters and 32B activated parameters, featuring a hybrid reasoning method that supports both thinking and direct response modes. Through multi-stage training on 23T tokens and comprehensive post-training with expert model iteration and reinforcement learning, GLM-4.5 achieves strong performance across agentic, reasoning, and coding (ARC) tasks, scoring 70.1% on TAU-Bench, 91.0% on AIME 24, and 64.2% on SWE-bench Verified. With much fewer parameters than several competitors, GLM-4.5 ranks 3rd overall among all evaluated models and 2nd on agentic benchmarks. We release both GLM-4.5 (355B parameters) and a compact version, GLM-4.5-Air (106B parameters), to advance research in reasoning and agentic AI systems. Code, models, and more information are available at https://github.com/zai-org/GLM-4.5.

Size Zheng, Xuegui Zheng, Li-wen Chang et al.

The exponential growth in Large Language Model (LLM) parameters has transformed model training into an increasingly resource-intensive endeavor. With the stagnation of Moore's Law and the widening disparity between computation throughput and communication bandwidth, expert parallelism (EP) has emerged as a critical strategy for scaling mixture-of-experts (MoE) models. However, despite numerous proposals for optimizing EP, ranging from communication compression to computation-communication overlap, adoption within production-grade frameworks like Megatron-LM remains conservative. Existing solutions often rely on ad-hoc, complex kernels that lack adaptability across diverse optimization configurations and frequently neglect numerical stability, failing to meet the strict precision requirements of large-scale training. In this paper, we introduce UniEP, a novel system that unifies diverse EP optimization strategies into a cohesive abstraction. UniEP fuses the MoE communication and computation into MegaKernels, effectively transforming complex architectural tuning into a unified parameter search space for automated adaptability. Crucially, UniEP incorporates a deterministic token ordering mechanism that guarantees numerical consistency with sequential execution, even under aggressive overlap schedules. We evaluate UniEP on GPU clusters equipped with NVIDIA Hopper GPUs. Our results demonstrate that UniEP achieves 1.03$\times$-1.38$\times$ speedups over state-of-the-art work, effectively mitigating communication bottlenecks while maintaining the rigorous accuracy standards required for production LLM training.

Hui Zeng, Daming Zhao, Pengfei Yang et al.

Generative reasoning with large language models (LLMs) often involves long decoding sequences, leading to substantial memory and latency overheads from accumulating key-value (KV) caches. While existing KV compression methods primarily focus on reducing prefill memory from long input sequences, they fall short in addressing the dynamic and layer-sensitive nature of long-form generation, which is central to reasoning tasks. We propose Lethe, a dynamic KV cache management framework that introduces adaptivity along both the spatial and temporal dimensions of decoding. Along the spatial dimension, Lethe performs layerwise sparsity-aware allocation, assigning token pruning budgets to each transformer layer based on estimated attention redundancy. Along the temporal dimension, Lethe conducts multi-round token pruning during generation, driven by a Recency-Aware Selective Retention} (RASR) mechanism. RASR extends traditional recency-based heuristics by also considering token relevance derived from evolving attention patterns, enabling informed decisions about which tokens to retain or evict. Empirical results demonstrate that Lethe achieves a favorable balance between efficiency and generation quality across diverse models and tasks, increases throughput by up to 2.56x.

Muyan Hu, Ashwin Venkatram, Shreyashri Biswas et al.

Kernel orchestration is the task of mapping the computation defined in different operators of a deep neural network (DNN) to the execution of GPU kernels on modern hardware platforms. Prior approaches optimize kernel orchestration by greedily applying operator fusion, which fuses the computation of multiple operators into a single kernel, and miss a variety of optimization opportunities in kernel orchestration. This paper presents Korch, a tensor program optimizer that discovers optimal kernel orchestration strategies for tensor programs. Instead of directly fusing operators, Korch first applies operator fission to decompose tensor operators into a small set of basic tensor algebra primitives. This decomposition enables a diversity of fine-grained, inter-operator optimizations. Next, Korch optimizes kernel orchestration by formalizing it as a constrained optimization problem, leveraging an off-the-shelf binary linear programming solver to discover an optimal orchestration strategy, and generating an executable that can be directly deployed on modern GPU platforms. Evaluation on a variety of DNNs shows that Korch outperforms existing tensor program optimizers by up to 1.7x on V100 GPUs and up to 1.6x on A100 GPUs. Korch is publicly available at https://github.com/humuyan/Korch.

Jiaao He, Jiezhong Qiu, Aohan Zeng et al.

Mixture-of-Expert (MoE) presents a strong potential in enlarging the size of language model to trillions of parameters. However, training trillion-scale MoE requires algorithm and system co-design for a well-tuned high performance distributed training system. Unfortunately, the only existing platform that meets the requirements strongly depends on Google's hardware (TPU) and software (Mesh Tensorflow) stack, and is not open and available to the public, especially GPU and PyTorch communities. In this paper, we present FastMoE, a distributed MoE training system based on PyTorch with common accelerators. The system provides a hierarchical interface for both flexible model design and easy adaption to different applications, such as Transformer-XL and Megatron-LM. Different from direct implementation of MoE models using PyTorch, the training speed is highly optimized in FastMoE by sophisticated high-performance acceleration skills. The system supports placing different experts on multiple GPUs across multiple nodes, enabling enlarging the number of experts linearly against the number of GPUs. The source of FastMoE is available at https://github.com/laekov/fastmoe under Apache-2 license.

Size Zheng, Xuegui Zheng, Hanshi Sun et al.

The scaling of large language models (LLMs) is currently bottlenecked by the rigidity of distributed programming. While high-performance libraries like CuBLAS and NCCL provide optimized primitives, they lack the flexibility required for rapidly evolving model architectures. Conversely, existing tensor compilers fail to address the complex memory hierarchy of distributed clusters effectively. To bridge this gap, we propose DITRON, a scalable tile-level compiler that democratizes high-performance distributed kernel development. DITRON introduces a novel hierarchical programming abstraction spanning Core, Device, and Task levels to map tensor programs efficiently onto heterogeneous distributed hardware. This abstraction allows DITRON to support diverse parallelism strategies while abstracting away the complexity of inter-node and intra-node communication. Evaluated across large-scale clusters, DITRON achieves performance parity with or exceeding expert-tuned CUDA libraries, delivering speedups of $6\%-30\%$ on isolated kernels and $5\%-30\%$ on end-to-end inference in vLLM. Furthermore, DITRON demonstrates strong portability, achieving significant speedups on both NVIDIA and AMD platforms. \ours{} has been deployed at the enterprise level for both training and inference. It achieves an MFU improvement of over 10\% in training tasks, saving approximately 500,000 GPU hours of training cost per month. For inference tasks, it delivers an end-to-end gain of over 20\% and has been applied to cloud service inference and edge inference scenarios.

Chia-chi Hsieh, Zan Zong, Xinyang Chen et al.

The growing demand for large language models (LLMs) requires serving systems to handle many concurrent requests with diverse service level objectives (SLOs). This exacerbates head-of-line (HoL) blocking during the compute-intensive prefill phase, where long-running requests monopolize resources and delay higher-priority ones, leading to widespread time-to-first-token (TTFT) SLO violations. While chunked prefill enables interruptibility, it introduces an inherent trade-off between responsiveness and throughput: reducing chunk size improves response latency but degrades computational efficiency, whereas increasing chunk size maximizes throughput but exacerbates blocking. This necessitates an adaptive preemption mechanism. However, dynamically balancing execution granularity against scheduling overheads remains a key challenge. In this paper, we propose FlowPrefill, a TTFT-goodput-optimized serving system that resolves this conflict by decoupling preemption granularity from scheduling frequency. To achieve adaptive prefill scheduling, FlowPrefill introduces two key innovations: 1) Operator-Level Preemption, which leverages operator boundaries to enable fine-grained execution interruption without the efficiency loss associated with fixed small chunking; and 2) Event-Driven Scheduling, which triggers scheduling decisions only upon request arrival or completion events, thereby supporting efficient preemption responsiveness while minimizing control-plane overhead. Evaluation on real-world production traces shows that FlowPrefill improves maximum goodput by up to 5.6$\times$ compared to state-of-the-art systems while satisfying heterogeneous SLOs.

Miao Tao, Yuanzhen Zhou, Haoran Xu et al.

Rendering large-scale 3D Gaussian Splatting (3DGS) model faces significant challenges in achieving real-time, high-fidelity performance on consumer-grade devices. Fully realizing the potential of 3DGS in applications such as virtual reality (VR) requires addressing critical system-level challenges to support real-time, immersive experiences. We propose GS-Cache, an end-to-end framework that seamlessly integrates 3DGS's advanced representation with a highly optimized rendering system. GS-Cache introduces a cache-centric pipeline to eliminate redundant computations, an efficiency-aware scheduler for elastic multi-GPU rendering, and optimized CUDA kernels to overcome computational bottlenecks. This synergy between 3DGS and system design enables GS-Cache to achieve up to 5.35x performance improvement, 35% latency reduction, and 42% lower GPU memory usage, supporting 2K binocular rendering at over 120 FPS with high visual quality. By bridging the gap between 3DGS's representation power and the demands of VR systems, GS-Cache establishes a scalable and efficient framework for real-time neural rendering in immersive environments.

Hang Wu, Jianian Zhu, Yinghui Li et al.

Large Language Models (LLMs) present a critical trade-off between inference quality and computational cost: larger models offer superior capabilities but incur significant latency, while smaller models are faster but less powerful. Existing serving strategies often employ fixed model scales or static two-stage speculative decoding, failing to dynamically adapt to the varying complexities of user requests or fluctuations in system performance. This paper introduces \systemname{}, a novel framework that reimagines LLM inference as an adaptive routing problem solved through multi-level speculative decoding. \systemname{} dynamically constructs and optimizes inference "paths" (chains of models) based on real-time feedback, addressing the limitations of static approaches. Our contributions are threefold: (1) An \textbf{adaptive model chain scheduling} mechanism that leverages performance profiling (execution times) and predictive similarity metrics (derived from token distribution divergence) to continuously select the optimal sequence of draft and verifier models, minimizing predicted latency per generated token. (2) A \textbf{multi-level collaborative verification} framework where intermediate models within the selected chain can validate speculative tokens, reducing the verification burden on the final, most powerful target model. (3) A \textbf{synchronized state management} system providing efficient, consistent KV cache handling across heterogeneous models in the chain, including precise, low-overhead rollbacks tailored for asynchronous batch processing inherent in multi-level speculation. Preliminary experiments demonstrate the validity of our method.

Zhixiang Ren, Yongheng Liu, Tianhui Shi et al.

The plethora of complex artificial intelligence (AI) algorithms and available high performance computing (HPC) power stimulates the expeditious development of AI components with heterogeneous designs. Consequently, the need for cross-stack performance benchmarking of AI-HPC systems emerges rapidly. The de facto HPC benchmark LINPACK can not reflect AI computing power and I/O performance without representative workload. The current popular AI benchmarks like MLPerf have fixed problem size therefore limited scalability. To address these issues, we propose an end-to-end benchmark suite utilizing automated machine learning (AutoML), which not only represents real AI scenarios, but also is auto-adaptively scalable to various scales of machines. We implement the algorithms in a highly parallel and flexible way to ensure the efficiency and optimization potential on diverse systems with customizable configurations. We utilize operations per second (OPS), which is measured in an analytical and systematic approach, as the major metric to quantify the AI performance. We perform evaluations on various systems to ensure the benchmark's stability and scalability, from 4 nodes with 32 NVIDIA Tesla T4 (56.1 Tera-OPS measured), up to 512 nodes with 4096 Huawei Ascend 910 (194.53 Peta-OPS measured), and the results show near-linear weak scalability. With flexible workload and single metric, our benchmark can scale and rank AI-HPC easily.