Apple Intelligence Foundation Language Models

We present foundation language models developed to power Apple Intelligence features, including a ~3 billion parameter model designed to run efficiently on devices and a large server-based language model designed for Private Cloud Compute. These models are designed to perform a wide range of tasks efficiently, accurately, and responsibly. This report describes the model architecture, the data used to train the model, the training process, how the models are optimized for inference, and the evaluation results. We highlight our focus on Responsible AI and how the principles are applied throughout the model development.

SeaLLMs 3: Open Foundation and Chat Multilingual Large Language Models for Southeast Asian Languages

Large Language Models (LLMs) have shown remarkable abilities across various tasks, yet their development has predominantly centered on high-resource languages like English and Chinese, leaving low-resource languages underserved. To address this disparity, we present SeaLLMs 3, the latest iteration of the SeaLLMs model family, tailored for Southeast Asian languages. This region, characterized by its rich linguistic diversity, has lacked adequate language technology support. SeaLLMs 3 aims to bridge this gap by covering a comprehensive range of languages spoken in this region, including English, Chinese, Indonesian, Vietnamese, Thai, Tagalog, Malay, Burmese, Khmer, Lao, Tamil, and Javanese. Leveraging efficient language enhancement techniques and a specially constructed instruction tuning dataset, SeaLLMs 3 significantly reduces training costs while maintaining high performance and versatility. Our model excels in tasks such as world knowledge, mathematical reasoning, translation, and instruction following, achieving state-of-the-art performance among similarly sized models. Additionally, we prioritized safety and reliability by addressing both general and culture-specific considerations and incorporated mechanisms to reduce hallucinations. This work underscores the importance of inclusive AI, showing that advanced LLM capabilities can benefit underserved linguistic and cultural communities.

Virtual Full-Duplex Wireless Communications with Zero-Interval Modulation and Sampling

In this paper, we propose a virtual full-duplex (VFD) technique with zero-interval modulation and sampling (ZIMS), where two half-duplex (HD) transceivers can simultaneously transmit signals and each transceiver can effectively receive the desired information. In ZIMS-VFD, the transceiver inserts a zero-interval for each symbol in the transmit signal and provides self-interference (SI)-free intervals for itself. Meanwhile, it samples the receive signal in the provided SI-free intervals and restores the desired symbols. Based on orthogonal frequency division multiplexing (OFDM), we formulate the system model and show the transmit signal structure. Then, we give the transceiver design for single input single output (SISO) ZIMS-VFD and extend it to multiple input multiple output (MIMO) communications. Numerical results verify our theoretical analyses and show that ZIMS-VFD can effectively increase the capacity and approach the FD without SI.

Multi-Frequency Resonant Circuit Based Multi-User Emergency Through-the-Earth Communication with Magnetic Induction

Magnetic induction (MI) is an effective technique in emergency through-the-earth communications due to the higher penetration efficiency and lower propagation loss as compared with electromagnetic wave communication. How to cancel the interference between different users and enhance the effectiveness of multi-user transmissions is imperative for the practical application of MI communication. In this paper, we use multi-frequency resonant circuit to establish multiple resonant frequencies for MI communication. The transmissions corresponding to different users operate at different resonant frequencies and multi-user interferences can be naturally mitigated. Numerical results verify our theoretical analyses and show that the proposed system can significantly enhance the performance.

RIS-Based Self-Interference Cancellation for Full-Duplex Broadband Transmission

Full-duplex (FD) is an attractive technology that can significantly boost the throughput of wireless communications. However, it is limited by the severe self-interference (SI) from the transmitter to the local receiver. In this paper, we propose a new SI cancellation (SIC) scheme based on reconfigurable intelligent surface (RIS), where small RISs are deployed inside FD devices to enhance SIC capability and system capacity under frequencyselective fading channels. The novel scheme can not only address the challenges associated with SIC but also improve the overall performance. We first analyze the near-field behavior of the RIS and then formulate an optimization problem to maximize the SIC capability by controlling the reflection coefficients (RCs) of the RIS and allocating the transmit power of the device. The problem is solved with alternate optimization (AO) algorithm in three cases: ideal case, where both the amplitude and phase of each RIS unit cell can be controlled independently and continuously, continuous phases, where the phase of each RIS unit cell can be controlled independently, while the amplitude is fixed to one, and discrete phases, where the RC of each RIS unit cell can only take discrete values and these discrete values are equally spaced on the unit circle. For the ideal case, the closed-form solution to RC is derived with Karush-Kuhn-Tucker (KKT) conditions. Based on Riemannian conjugate gradient (RCG) algorithm, we optimize the RC for the case of continuous phases and then extend the solution to the case of discrete phases by the nearest point projection (NPP) method. Simulation results are given to validate the performance of our proposed SIC scheme.

Fast and Slow Generating: An Empirical Study on Large and Small Language Models Collaborative Decoding

Large Language Models (LLMs) demonstrate impressive performance in diverse applications, yet they face significant drawbacks, including high inference latency, expensive training cost, and generation of hallucination. Collaborative decoding between large and small language models (SLMs) offers a novel approach to address these challenges. Inspired by dual-process cognitive theory, we integrate these methods into a unified framework termed Fast and Slow Generating (FS-GEN). This paper explores several techniques within the FS-GEN framework, including speculative decoding, contrastive decoding, and emulator or proxy fine-tuning. We provide a comprehensive analysis of these methodologies, offering insights into their similarities and differences under this framework. Our study delves into the differential knowledge capabilities of LLMs versus SLMs through the FS-GEN lens, revealing that fewer than 20% of collaborative interactions are required across various methods. These interactions adhere to a scaling law relative to the parameter ratios, thereby facilitating predictable collaboration. Furthermore, we investigate the specific positions where collaboration is most effective from an uncertainty perspective, yielding novel insights that could refine FS-GEN methods. Our findings reveal that the essential difference between models of different sizes lies in the uncertainty of the next token prediction, where interventions by larger models are most needed to assist the smaller ones. Code for Reproduction: https://github.com/TsinghuaC3I/FS-GEN

Determined Multichannel Blind Source Separation with Clustered Source Model

The independent low-rank matrix analysis (ILRMA) method stands out as a prominent technique for multichannel blind audio source separation. It leverages nonnegative matrix factorization (NMF) and nonnegative canonical polyadic decomposition (NCPD) to model source parameters. While it effectively captures the low-rank structure of sources, the NMF model overlooks inter-channel dependencies. On the other hand, NCPD preserves intrinsic structure but lacks interpretable latent factors, making it challenging to incorporate prior information as constraints. To address these limitations, we introduce a clustered source model based on nonnegative block-term decomposition (NBTD). This model defines blocks as outer products of vectors (clusters) and matrices (for spectral structure modeling), offering interpretable latent vectors. Moreover, it enables straightforward integration of orthogonality constraints to ensure independence among source images. Experimental results demonstrate that our proposed method outperforms ILRMA and its extensions in anechoic conditions and surpasses the original ILRMA in simulated reverberant environments.

MM1: Methods, Analysis & Insights from Multimodal LLM Pre-training

In this work, we discuss building performant Multimodal Large Language Models (MLLMs). In particular, we study the importance of various architecture components and data choices. Through careful and comprehensive ablations of the image encoder, the vision language connector, and various pre-training data choices, we identified several crucial design lessons. For example, we demonstrate that for large-scale multimodal pre-training using a careful mix of image-caption, interleaved image-text, and text-only data is crucial for achieving state-of-the-art (SOTA) few-shot results across multiple benchmarks, compared to other published pre-training results. Further, we show that the image encoder together with image resolution and the image token count has substantial impact, while the vision-language connector design is of comparatively negligible importance. By scaling up the presented recipe, we build MM1, a family of multimodal models up to 30B parameters, including both dense models and mixture-of-experts (MoE) variants, that are SOTA in pre-training metrics and achieve competitive performance after supervised fine-tuning on a range of established multimodal benchmarks. Thanks to large-scale pre-training, MM1 enjoys appealing properties such as enhanced in-context learning, and multi-image reasoning, enabling few-shot chain-of-thought prompting.

CoGenesis: A Framework Collaborating Large and Small Language Models for Secure Context-Aware Instruction Following

With the advancement of language models (LMs), their exposure to private data is increasingly inevitable, and their deployment (especially for smaller ones) on personal devices, such as PCs and smartphones, has become a prevailing trend. In contexts laden with user information, enabling models to both safeguard user privacy and execute commands efficiently emerges as an essential research imperative. In this paper, we propose CoGenesis, a collaborative generation framework integrating large (hosted on cloud infrastructure) and small models (deployed on local devices) to address privacy concerns logically. Initially, we design a pipeline to create personalized writing instruction datasets enriched with extensive context details as the testbed of this research issue. Subsequently, we introduce two variants of CoGenesis based on sketch and logits respectively. Our experimental findings, based on our synthesized dataset and two additional open-source datasets, indicate that: 1) Large-scale models perform well when provided with user context but struggle in the absence of such context. 2) While specialized smaller models fine-tuned on the synthetic dataset show promise, they still lag behind their larger counterparts. 3) Our CoGenesis framework, utilizing mixed-scale models, showcases competitive performance, providing a feasible solution to privacy issues.

Momentum Approximation in Asynchronous Private Federated Learning

Asynchronous protocols have been shown to improve the scalability of federated learning (FL) with a massive number of clients. Meanwhile, momentum-based methods can achieve the best model quality in synchronous FL. However, naively applying momentum in asynchronous FL algorithms leads to slower convergence and degraded model performance. It is still unclear how to effective combinie these two techniques together to achieve a win-win. In this paper, we find that asynchrony introduces implicit bias to momentum updates. In order to address this problem, we propose momentum approximation that minimizes the bias by finding an optimal weighted average of all historical model updates. Momentum approximation is compatible with secure aggregation as well as differential privacy, and can be easily integrated in production FL systems with a minor communication and storage cost. We empirically demonstrate that on benchmark FL datasets, momentum approximation can achieve $1.15 \textrm{--}4\times$ speed up in convergence compared to existing asynchronous FL optimizers with momentum.