Weilin Zhao

Shengding Hu, Ning Ding, Weilin Zhao et al. · tsinghua

The scale of large pre-trained models (PTMs) poses significant challenges in adapting to downstream tasks due to the high optimization overhead and storage costs associated with full-parameter fine-tuning. To address this, many studies explore parameter-efficient tuning methods, also framed as "delta tuning", which updates only a small subset of parameters, known as "delta modules", while keeping the backbone model's parameters fixed. However, the practicality and flexibility of delta tuning have been limited due to existing implementations that directly modify the code of the backbone PTMs and hard-code specific delta tuning methods for each PTM. In this paper, we present OpenDelta, an open-source library that overcomes these limitations by providing a plug-and-play implementation of various delta tuning methods. Our novel techniques eliminate the need to modify the backbone PTMs' code, making OpenDelta compatible with different, even novel PTMs. OpenDelta is designed to be simple, modular, and extensible, providing a comprehensive platform for researchers and practitioners to adapt large PTMs efficiently.

Weilin Zhao, Yuxiang Huang, Xu Han et al. · tsinghua

Recently, there has been a demand to deploy Large Language Models (LLMs) on personal devices such as laptops and smartphones. These LLMs have different model variants when handling different tasks. However, personal devices have limited resources and require reduced storage overhead. To address this, there are two key methods available: the first is model compression, which compresses LLMs into smaller sizes; the second is LoRA, which can transfer an LLM to other tasks with very few parameters, avoiding the storage of multiple model variants in multi-task scenarios by only preserving LoRAs. However, our experiments show that directly combining these two methods yields sub-optimal performance. Considering that the open-source community has already contributed many LoRAs to LLMs, we propose to adapt these existing LoRAs from the LLMs to their compressed version and introduce a Compression-Aware LoRA (CA-LoRA) framework. We incorporate knowledge inheritance and recovery strategies to recover the lost knowledge caused by model compression. Experiment results demonstrate that CA-LoRA outperforms the vanilla LoRA methods applied to a compressed LLM and achieves comparable performance to the non-compressed LLM with existing LoRA modules. The source code of CA-LoRA is available at https://github.com/thunlp/CA-LoRA.

Yujia Qin, Shengding Hu, Yankai Lin et al. · tsinghua

Humans possess an extraordinary ability to create and utilize tools, allowing them to overcome physical limitations and explore new frontiers. With the advent of foundation models, AI systems have the potential to be equally adept in tool use as humans. This paradigm, i.e., tool learning with foundation models, combines the strengths of specialized tools and foundation models to achieve enhanced accuracy, efficiency, and automation in problem-solving. Despite its immense potential, there is still a lack of a comprehensive understanding of key challenges, opportunities, and future endeavors in this field. To this end, we present a systematic investigation of tool learning in this paper. We first introduce the background of tool learning, including its cognitive origins, the paradigm shift of foundation models, and the complementary roles of tools and models. Then we recapitulate existing tool learning research into tool-augmented and tool-oriented learning. We formulate a general tool learning framework: starting from understanding the user instruction, models should learn to decompose a complex task into several subtasks, dynamically adjust their plan through reasoning, and effectively conquer each sub-task by selecting appropriate tools. We also discuss how to train models for improved tool-use capabilities and facilitate the generalization in tool learning. Considering the lack of a systematic tool learning evaluation in prior works, we experiment with 18 representative tools and show the potential of current foundation models in skillfully utilizing tools. Finally, we discuss several open problems that require further investigation for tool learning. In general, we hope this paper could inspire future research in integrating tools with foundation models.

Ning Ding, Yujia Qin, Guang Yang et al. · tsinghua

Despite the success, the process of fine-tuning large-scale PLMs brings prohibitive adaptation costs. In fact, fine-tuning all the parameters of a colossal model and retaining separate instances for different tasks are practically infeasible. This necessitates a new branch of research focusing on the parameter-efficient adaptation of PLMs, dubbed as delta tuning in this paper. In contrast with the standard fine-tuning, delta tuning only fine-tunes a small portion of the model parameters while keeping the rest untouched, largely reducing both the computation and storage costs. Recent studies have demonstrated that a series of delta tuning methods with distinct tuned parameter selection could achieve performance on a par with full-parameter fine-tuning, suggesting a new promising way of stimulating large-scale PLMs. In this paper, we first formally describe the problem of delta tuning and then comprehensively review recent delta tuning approaches. We also propose a unified categorization criterion that divide existing delta tuning methods into three groups: addition-based, specification-based, and reparameterization-based methods. Though initially proposed as an efficient method to steer large models, we believe that some of the fascinating evidence discovered along with delta tuning could help further reveal the mechanisms of PLMs and even deep neural networks. To this end, we discuss the theoretical principles underlying the effectiveness of delta tuning and propose frameworks to interpret delta tuning from the perspective of optimization and optimal control, respectively. Furthermore, we provide a holistic empirical study of representative methods, where results on over 100 NLP tasks demonstrate a comprehensive performance comparison of different approaches. The experimental results also cover the analysis of combinatorial, scaling and transferable properties of delta tuning.

Yuan Yao, Tianyu Yu, Ao Zhang et al. · tsinghua

The recent surge of Multimodal Large Language Models (MLLMs) has fundamentally reshaped the landscape of AI research and industry, shedding light on a promising path toward the next AI milestone. However, significant challenges remain preventing MLLMs from being practical in real-world applications. The most notable challenge comes from the huge cost of running an MLLM with a massive number of parameters and extensive computation. As a result, most MLLMs need to be deployed on high-performing cloud servers, which greatly limits their application scopes such as mobile, offline, energy-sensitive, and privacy-protective scenarios. In this work, we present MiniCPM-V, a series of efficient MLLMs deployable on end-side devices. By integrating the latest MLLM techniques in architecture, pretraining and alignment, the latest MiniCPM-Llama3-V 2.5 has several notable features: (1) Strong performance, outperforming GPT-4V-1106, Gemini Pro and Claude 3 on OpenCompass, a comprehensive evaluation over 11 popular benchmarks, (2) strong OCR capability and 1.8M pixel high-resolution image perception at any aspect ratio, (3) trustworthy behavior with low hallucination rates, (4) multilingual support for 30+ languages, and (5) efficient deployment on mobile phones. More importantly, MiniCPM-V can be viewed as a representative example of a promising trend: The model sizes for achieving usable (e.g., GPT-4V) level performance are rapidly decreasing, along with the fast growth of end-side computation capacity. This jointly shows that GPT-4V level MLLMs deployed on end devices are becoming increasingly possible, unlocking a wider spectrum of real-world AI applications in the near future.

Shengding Hu, Xin Liu, Xu Han et al. · tencent-ai, tsinghua

The scientific scale-up of large language models (LLMs) necessitates a comprehensive understanding of their scaling properties. However, the existing literature on the scaling properties only yields an incomplete answer: optimization loss decreases predictably as the model size increases, in line with established scaling law; yet no scaling law for task has been established and the task performances are far from predictable during scaling. Task performances typically show minor gains on small models until they improve dramatically once models exceed a size threshold, exemplifying the ``emergent abilities''. In this study, we discover that small models, although they exhibit minor performance, demonstrate critical and consistent task performance improvements that are not captured by conventional evaluation strategies due to insufficient measurement resolution. To measure such improvements, we introduce PassUntil, an evaluation strategy with theoretically infinite resolution, through massive sampling in the decoding phase. With PassUntil, we conduct a quantitative investigation into the scaling law of task performance. The investigation contains two parts. Firstly, a strict task scaling law that is not conventionally known to exist, is identified, enhancing the predictability of task performances. Remarkably, we are able to predict the performance of the 2.4B model on code generation with merely 0.05\% deviation before training starts, which is the first systematic attempt to verify predictable scaling proposed by GPT-4's report. Secondly, we are able to study emergent abilities quantitatively. We identify a kind of accelerated emergence whose scaling curve cannot be fitted by standard scaling law function and has a increasing speed. We then examine two hypothesis and imply that the ``multiple circuits hypothesis'' might be responsible for the accelerated emergence.

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.

Chaojun Xiao, Zhengyan Zhang, Chenyang Song et al. · tencent-ai, tsinghua

Advancements in LLMs have recently unveiled challenges tied to computational efficiency and continual scalability due to their requirements of huge parameters, making the applications and evolution of these models on devices with limited computation resources and scenarios requiring various abilities increasingly cumbersome. Inspired by modularity within the human brain, there is a growing tendency to decompose LLMs into numerous functional modules, allowing for inference with part of modules and dynamic assembly of modules to tackle complex tasks, such as mixture-of-experts. To highlight the inherent efficiency and composability of the modular approach, we coin the term brick to represent each functional module, designating the modularized structure as configurable foundation models. In this paper, we offer a comprehensive overview and investigation of the construction, utilization, and limitation of configurable foundation models. We first formalize modules into emergent bricks - functional neuron partitions that emerge during the pre-training phase, and customized bricks - bricks constructed via additional post-training to improve the capabilities and knowledge of LLMs. Based on diverse functional bricks, we further present four brick-oriented operations: retrieval and routing, merging, updating, and growing. These operations allow for dynamic configuration of LLMs based on instructions to handle complex tasks. To verify our perspective, we conduct an empirical analysis on widely-used LLMs. We find that the FFN layers follow modular patterns with functional specialization of neurons and functional neuron partitions. Finally, we highlight several open issues and directions for future research. Overall, this paper aims to offer a fresh modular perspective on existing LLM research and inspire the future creation of more efficient and scalable foundational models.

Wang Xu, Shuo Wang, Weilin Zhao et al. · tsinghua

Large language models (LLMs) have demonstrated the ability to improve human efficiency through conversational interactions. Conventional LLM-powered dialogue systems, operating on a turn-based paradigm, preclude real-time interaction during response generation. To address this limitation, researchers have proposed duplex models. These models can dynamically adapt to user input, facilitating real-time interactive feedback. However, these methods typically require substantial computational resources to acquire the ability. To reduce overhead, this paper presents a new duplex decoding approach that enhances LLMs with duplex ability, requiring minimal additional training. Specifically, our method employs parallel decoding of queries and responses in conversations, effectively implementing a channel-division-multiplexing decoding strategy. Experimental results indicate that our proposed method significantly enhances the naturalness and human-likeness of user-AI interactions with minimal training costs.

Yuxiang Huang, Mingye Li, Xu Han et al.

The efficiency of long-video inference remains a critical bottleneck, mainly due to the dense computation in the prefill stage of Large Multimodal Models (LMMs). Existing methods either compress visual embeddings or apply sparse attention on a single GPU, yielding limited acceleration or degraded performance and restricting LMMs from handling longer, more complex videos. To overcome these issues, we propose Spava, a sequence-parallel framework with optimized attention that accelerates long-video inference across multiple GPUs. By distributing approximate attention, Spava reduces computation and increases parallelism, enabling efficient processing of more visual embeddings without compression and thereby improving task performance. System-level optimizations, such as load balancing and fused forward passes, further unleash the potential of Spava, delivering speedups of 12.72x, 1.70x, and 1.18x over FlashAttn, ZigZagRing, and APB, without notable performance loss. Code available at https://github.com/thunlp/APB

Shengding Hu, Yuge Tu, Xu Han et al. · tsinghua

The burgeoning interest in developing Large Language Models (LLMs) with up to trillion parameters has been met with concerns regarding resource efficiency and practical expense, particularly given the immense cost of experimentation. This scenario underscores the importance of exploring the potential of Small Language Models (SLMs) as a resource-efficient alternative. In this context, we introduce MiniCPM, specifically the 1.2B and 2.4B non-embedding parameter variants, not only excel in their respective categories but also demonstrate capabilities on par with 7B-13B LLMs. While focusing on SLMs, our approach exhibits scalability in both model and data dimensions for future LLM research. Regarding model scaling, we employ extensive model wind tunnel experiments for stable and optimal scaling. For data scaling, we introduce a Warmup-Stable-Decay (WSD) learning rate scheduler (LRS), conducive to continuous training and domain adaptation. We present an in-depth analysis of the intriguing training dynamics that occurred in the WSD LRS. With WSD LRS, we are now able to efficiently study data-model scaling law without extensive retraining experiments on both axes of model and data, from which we derive the much higher compute optimal data-model ratio than Chinchilla Optimal. Additionally, we introduce MiniCPM family, including MiniCPM-DPO, MiniCPM-MoE and MiniCPM-128K, whose excellent performance further cementing MiniCPM's foundation in diverse SLM applications. MiniCPM models are available publicly at https://github.com/OpenBMB/MiniCPM .

Zanlin Ni, Shenzhi Wang, Yang Yue et al.

Diffusion Large Language Models (dLLMs) break the rigid left-to-right constraint of traditional LLMs, enabling token generation in arbitrary orders. Intuitively, this flexibility implies a solution space that strictly supersets the fixed autoregressive trajectory, theoretically unlocking superior reasoning potential for general tasks like mathematics and coding. Consequently, numerous works have leveraged reinforcement learning (RL) to elicit the reasoning capability of dLLMs. In this paper, we reveal a counter-intuitive reality: arbitrary order generation, in its current form, narrows rather than expands the reasoning boundary of dLLMs. We find that dLLMs tend to exploit this order flexibility to bypass high-uncertainty tokens that are crucial for exploration, leading to a premature collapse of the solution space. This observation challenges the premise of existing RL approaches for dLLMs, where considerable complexities, such as handling combinatorial trajectories and intractable likelihoods, are often devoted to preserving this flexibility. We demonstrate that effective reasoning is better elicited by intentionally forgoing arbitrary order and applying standard Group Relative Policy Optimization (GRPO) instead. Our approach, JustGRPO, is minimalist yet surprisingly effective (e.g., 89.1% accuracy on GSM8K) while fully retaining the parallel decoding ability of dLLMs. Project page: https://nzl-thu.github.io/the-flexibility-trap

Weilin Zhao, Yuxiang Huang, Xu Han et al. · tsinghua

Speculative decoding is a widely used method that accelerates the generation process of large language models (LLMs) with no compromise in model performance. It achieves this goal by using an existing smaller model for drafting and then employing the target LLM to verify the draft in a low-cost parallel manner. Under such a drafting-verification framework, drafting efficiency has become a bottleneck in the final speedup of speculative decoding. Therefore, generating longer drafts at less cost can lead to better decoding speedup. To achieve this, we introduce Ouroboros, which can generate draft phrases to parallelize the drafting process and meanwhile lengthen drafts in a training-free manner. The experimental results on various typical text generation tasks show that Ouroboros can achieve speedups of up to $2.8\times$ over speculative decoding and $3.9\times$ over vanilla decoding, without fine-tuning draft and target models. The source code of Ouroboros is available at https://github.com/thunlp/Ouroboros.

Chaojun Xiao, Jie Cai, Weilin Zhao et al. · tsinghua

Large Language Models (LLMs) have emerged as a milestone in artificial intelligence, and their performance can improve as the model size increases. However, this scaling brings great challenges to training and inference efficiency, particularly for deploying LLMs in resource-constrained environments, and the scaling trend is becoming increasingly unsustainable. This paper introduces the concept of ``\textit{capacity density}'' as a new metric to evaluate the quality of the LLMs across different scales and describes the trend of LLMs in terms of both effectiveness and efficiency. To calculate the capacity density of a given target LLM, we first introduce a set of reference models and develop a scaling law to predict the downstream performance of these reference models based on their parameter sizes. We then define the \textit{effective parameter size} of the target LLM as the parameter size required by a reference model to achieve equivalent performance, and formalize the capacity density as the ratio of the effective parameter size to the actual parameter size of the target LLM. Capacity density provides a unified framework for assessing both model effectiveness and efficiency. Our further analysis of recent open-source base LLMs reveals an empirical law (the densing law)that the capacity density of LLMs grows exponentially over time. More specifically, using some widely used benchmarks for evaluation, the capacity density of LLMs doubles approximately every three months. The law provides new perspectives to guide future LLM development, emphasizing the importance of improving capacity density to achieve optimal results with minimal computational overhead.

MiniCPM Team, Chaojun Xiao, Yuxuan Li et al. · tencent-ai, tsinghua

This paper introduces MiniCPM4, a highly efficient large language model (LLM) designed explicitly for end-side devices. We achieve this efficiency through systematic innovation in four key dimensions: model architecture, training data, training algorithms, and inference systems. Specifically, in terms of model architecture, we propose InfLLM v2, a trainable sparse attention mechanism that accelerates both prefilling and decoding phases for long-context processing. Regarding training data, we propose UltraClean, an efficient and accurate pre-training data filtering and generation strategy, and UltraChat v2, a comprehensive supervised fine-tuning dataset. These datasets enable satisfactory model performance to be achieved using just 8 trillion training tokens. Regarding training algorithms, we propose ModelTunnel v2 for efficient pre-training strategy search, and improve existing post-training methods by introducing chunk-wise rollout for load-balanced reinforcement learning and data-efficient tenary LLM, BitCPM. Regarding inference systems, we propose CPM.cu that integrates sparse attention, model quantization, and speculative sampling to achieve efficient prefilling and decoding. To meet diverse on-device requirements, MiniCPM4 is available in two versions, with 0.5B and 8B parameters, respectively. Furthermore, we construct a hybrid reasoning model, MiniCPM4.1, which can be used in both deep reasoning mode and non-reasoning mode. Evaluation results demonstrate that MiniCPM4 and MiniCPM4.1 outperform similar-sized open-source models across benchmarks, with the 8B variants showing significant speed improvements on long sequence understanding and generation.

Weilin Zhao, Tengyu Pan, Xu Han et al. · tsinghua

Speculative sampling has emerged as an important technique for accelerating the auto-regressive generation process of large language models (LLMs) by utilizing a draft-then-verify mechanism to produce multiple tokens per forward pass. While state-of-the-art speculative sampling methods use only a single layer and a language modeling (LM) head as the draft model to achieve impressive layer compression, their efficiency gains are substantially reduced for large-vocabulary LLMs, such as Llama-3-8B with a vocabulary of 128k tokens. To address this, we present FR-Spec, a frequency-ranked speculative sampling framework that optimizes draft candidate selection through vocabulary space compression. By constraining the draft search to a frequency-prioritized token subset, our method reduces LM Head computation overhead by 75% while ensuring the equivalence of the final output distribution. Experiments across multiple datasets demonstrate an average of 1.12$\times$ speedup over the state-of-the-art speculative sampling method EAGLE-2. Code available at https://github.com/thunlp/FR-Spec.

Weilin Zhao, Zihan Zhou, Zhou Su et al. · tsinghua

Long-sequence processing is a critical capability for modern large language models. However, the self-attention mechanism in the standard Transformer architecture faces severe computational and memory bottlenecks when processing long sequences. While trainable sparse attention methods offer a promising solution, existing approaches such as NSA introduce excessive extra parameters and disrupt the conventional \textit{pretrain-on-short, finetune-on-long} workflow, resulting in slow convergence and difficulty in acceleration. To overcome these limitations, we introduce dense-sparse switchable attention framework, termed as InfLLM-V2. InfLLM-V2 is a trainable sparse attention that seamlessly adapts models from short to long sequences. Specifically, InfLLM-V2 reuses dense attention parameters through parameter-free architecture modification, maintaining consistency between short and long sequence processing. Additionally, InfLLM-V2 ensures computational efficiency across all sequence lengths, by using dense attention for short inputs and smoothly transitioning to sparse attention for long sequences. To achieve practical acceleration, we further introduce an efficient implementation of InfLLM-V2 that significantly reduces the computational overhead. Our experiments on long-context understanding and chain-of-thought reasoning demonstrate that InfLLM-V2 is 4$\times$ faster than dense attention while retaining 98.1% and 99.7% of the performance, respectively. Based on the InfLLM-V2 framework, we have trained and open-sourced MiniCPM4.1 (https://huggingface.co/openbmb/MiniCPM4.1-8B), a hybrid reasoning model, providing a reproducible implementation for the research community.

Yudi Zhang, Weilin Zhao, Xu Han et al. · tsinghua

Speculative decoding and quantization effectively accelerate memory-bound inference of large language models. Speculative decoding mitigates the memory bandwidth bottleneck by verifying multiple tokens within a single forward pass, which increases computational effort. Quantization achieves this optimization by compressing weights and activations into lower bit-widths and also reduces computations via low-bit matrix multiplications. To further leverage their strengths, we investigate the integration of these two techniques. Surprisingly, experiments applying the advanced speculative decoding method EAGLE-2 to various quantized models reveal that the memory benefits from 4-bit weight quantization are diminished by the computational load from speculative decoding. Specifically, verifying a tree-style draft incurs significantly more time overhead than a single-token forward pass on 4-bit weight quantized models. This finding led to our new speculative decoding design: a hierarchical framework that employs a small model as an intermediate stage to turn tree-style drafts into sequence drafts, leveraging the memory access benefits of the target quantized model. Experimental results show that our hierarchical approach achieves a 2.78$\times$ speedup across various tasks for the 4-bit weight Llama-3-70B model on an A100 GPU, outperforming EAGLE-2 by 1.31$\times$. Code available at https://github.com/AI9Stars/SpecMQuant.

Yuxiang Huang, Mingye Li, Xu Han et al. · tsinghua

While long-context inference is crucial for advancing large language model (LLM) applications, its prefill speed remains a significant bottleneck. Current approaches, including sequence parallelism strategies and compute reduction through approximate attention mechanisms, still fall short of delivering optimal inference efficiency. This hinders scaling the inputs to longer sequences and processing long-context queries in a timely manner. To address this, we introduce APB, an efficient long-context inference framework that leverages multi-host approximate attention to enhance prefill speed by reducing compute and enhancing parallelism simultaneously. APB introduces a communication mechanism for essential key-value pairs within a sequence parallelism framework, enabling a faster inference speed while maintaining task performance. We implement APB by incorporating a tailored FlashAttn kernel alongside optimized distribution strategies, supporting diverse models and parallelism configurations. APB achieves speedups of up to 9.2x, 4.2x, and 1.6x compared with FlashAttn, RingAttn, and StarAttn, respectively, without any observable task performance degradation. We provide the implementation and experiment code of APB in https://github.com/thunlp/APB.

Chenyang Song, Weilin Zhao, Xu Han et al.

While Mixture-of-Experts (MoE) scales model capacity without proportionally increasing computation, its massive total parameter footprint creates significant storage and memory-access bottlenecks, which hinder efficient end-side deployment that simultaneously requires high performance, low computational cost, and small storage overhead. To achieve these properties, we present DECO, a sparse MoE architecture designed to match the performance of dense Transformers under identical total parameter budgets and training tokens. DECO utilizes the differentiable and flexible ReLU-based routing enhanced by learnable expert-wise scaling, which adaptively balances the contributions of routed and shared experts. Furthermore, we introduce NormSiLU, an activation function that normalizes inputs prior to SiLU operators, producing a more stable trend of routed-expert activation ratio and a higher intrinsic sparsity level. We also identify an empirical advantage in using non-gated MLP experts with ReLU-based routing, indicating the possibility of MoE architecture simplification. Experiments demonstrate that DECO, activating only 20% of experts, matches dense performance and outperforms established MoE baselines. Our specialized acceleration kernel delivers a 3.00$\times$ speedup on real hardware compared with dense inference. Codes and checkpoints will be released.

Chenyang Song, Weilin Zhao, Xu Han et al. · tsinghua

To alleviate the computational burden of large language models (LLMs), architectures with activation sparsity, represented by mixture-of-experts (MoE), have attracted increasing attention. However, the non-differentiable and inflexible routing of vanilla MoE hurts model performance. Moreover, while each token activates only a few parameters, these sparsely-activated architectures exhibit low chunk-level sparsity, indicating that the union of multiple consecutive tokens activates a large ratio of parameters. Such a sparsity pattern is unfriendly for acceleration under low-resource conditions (e.g., end-side devices) and incompatible with mainstream acceleration techniques (e.g., speculative decoding). To address these challenges, we introduce a novel MoE architecture, BlockFFN, as well as its efficient training and deployment techniques. Specifically, we use a router integrating ReLU activation and RMSNorm for differentiable and flexible routing. Next, to promote both token-level sparsity (TLS) and chunk-level sparsity (CLS), CLS-aware training objectives are designed, making BlockFFN more acceleration-friendly. Finally, we implement efficient acceleration kernels, combining activation sparsity and speculative decoding for the first time. The experimental results demonstrate the superior performance of BlockFFN over other MoE baselines, achieving over 80% TLS and 70% 8-token CLS. Our kernels achieve up to 3.67$\times$ speedup on real end-side devices than dense models. All codes and checkpoints are available publicly (https://github.com/thunlp/BlockFFN).

Ning Ding, Shengding Hu, Weilin Zhao et al.

Prompt-learning has become a new paradigm in modern natural language processing, which directly adapts pre-trained language models (PLMs) to $cloze$-style prediction, autoregressive modeling, or sequence to sequence generation, resulting in promising performances on various tasks. However, no standard implementation framework of prompt-learning is proposed yet, and most existing prompt-learning codebases, often unregulated, only provide limited implementations for specific scenarios. Since there are many details such as templating strategy, initializing strategy, and verbalizing strategy, etc. need to be considered in prompt-learning, practitioners face impediments to quickly adapting the desired prompt learning methods to their applications. In this paper, we present {OpenPrompt}, a unified easy-to-use toolkit to conduct prompt-learning over PLMs. OpenPrompt is a research-friendly framework that is equipped with efficiency, modularity, and extendibility, and its combinability allows the freedom to combine different PLMs, task formats, and prompting modules in a unified paradigm. Users could expediently deploy prompt-learning frameworks and evaluate the generalization of them on different NLP tasks without constraints. OpenPrompt is publicly released at {\url{ https://github.com/thunlp/OpenPrompt}}.

Ning Ding, Yulin Chen, Ganqu Cui et al. · tsinghua

Underlying data distributions of natural language, programming code, and mathematical symbols vary vastly, presenting a complex challenge for large language models (LLMs) that strive to achieve high performance across all three domains simultaneously. Achieving a very high level of proficiency for an LLM within a specific domain often requires extensive training with relevant corpora, which is typically accompanied by a sacrifice in performance in other domains. In this paper, we propose to fuse models that are already highly-specialized directly. The proposed fusing framework, UltraFuser, consists of three distinct specialists that are already sufficiently trained on language, coding, and mathematics. A token-level gating mechanism is introduced to blend the specialists' outputs. A two-stage training strategy accompanied by balanced sampling is designed to ensure stability. To effectively train the fused model, we further construct a high-quality supervised instruction tuning dataset, UltraChat 2, which includes text, code, and mathematical content. This dataset comprises approximately 300,000 instructions and covers a wide range of topics in each domain. Experiments show that our model could simultaneously achieve mastery of the three crucial domains.

Xinrong Zhang, Yingfa Chen, Shengding Hu et al.

As large language models (LLMs) increasingly permeate daily lives, there is a growing demand for real-time interactions that mirror human conversations. Traditional turn-based chat systems driven by LLMs prevent users from verbally interacting with the system while it is generating responses. To overcome these limitations, we adapt existing LLMs to \textit{duplex models} so that these LLMs can listen for users while generating output and dynamically adjust themselves to provide users with instant feedback. % such as in response to interruptions. Specifically, we divide the queries and responses of conversations into several time slices and then adopt a time-division-multiplexing (TDM) encoding-decoding strategy to pseudo-simultaneously process these slices. Furthermore, to make LLMs proficient enough to handle real-time conversations, we build a fine-tuning dataset consisting of alternating time slices of queries and responses as well as covering typical feedback types in instantaneous interactions. Our experiments show that although the queries and responses of conversations are segmented into incomplete slices for processing, LLMs can preserve their original performance on standard benchmarks with a few fine-tuning steps on our dataset. Automatic and human evaluation indicate that duplex models make user-AI interactions more natural and human-like, and greatly improve user satisfaction compared to vanilla LLMs. Our duplex model and dataset will be released.

Ao Sun, Weilin Zhao, Xu Han et al.

Effective attention modules have played a crucial role in the success of Transformer-based large language models (LLMs), but the quadratic time and memory complexities of these attention modules also pose a challenge when processing long sequences. One potential solution for the long sequence problem is to utilize distributed clusters to parallelize the computation of attention modules across multiple devices (e.g., GPUs). However, adopting a distributed approach inevitably introduces extra memory overheads to store local attention results and incurs additional communication costs to aggregate local results into global ones. In this paper, we propose a distributed attention framework named ``BurstAttention'' to optimize memory access and communication operations at both the global cluster and local device levels. In our experiments, we compare BurstAttention with other competitive distributed attention solutions for long sequence processing. The experimental results under different length settings demonstrate that BurstAttention offers significant advantages for processing long sequences compared with these competitive baselines, reducing 40% communication overheads and achieving 1.37 X speedup during training 128K sequence length on 32 X A100.

Xu Han, Weilin Zhao, Ning Ding et al.

Fine-tuned pre-trained language models (PLMs) have achieved awesome performance on almost all NLP tasks. By using additional prompts to fine-tune PLMs, we can further stimulate the rich knowledge distributed in PLMs to better serve downstream tasks. Prompt tuning has achieved promising results on some few-class classification tasks such as sentiment classification and natural language inference. However, manually designing lots of language prompts is cumbersome and fallible. For those auto-generated prompts, it is also expensive and time-consuming to verify their effectiveness in non-few-shot scenarios. Hence, it is still challenging for prompt tuning to address many-class classification tasks. To this end, we propose prompt tuning with rules (PTR) for many-class text classification and apply logic rules to construct prompts with several sub-prompts. In this way, PTR is able to encode prior knowledge of each class into prompt tuning. We conduct experiments on relation classification, a typical and complicated many-class classification task, and the results show that PTR can significantly and consistently outperform existing state-of-the-art baselines. This indicates that PTR is a promising approach to take advantage of both human prior knowledge and PLMs for those complicated classification tasks.