Lifelong Language Pretraining with Distribution-Specialized Experts

Pretraining on a large-scale corpus has become a standard method to build general language models (LMs). Adapting a model to new data distributions targeting different downstream tasks poses significant challenges. Naive fine-tuning may incur catastrophic forgetting when the over-parameterized LMs overfit the new data but fail to preserve the pretrained features. Lifelong learning (LLL) aims to enable information systems to learn from a continuous data stream across time. However, most prior work modifies the training recipe assuming a static fixed network architecture. We find that additional model capacity and proper regularization are key elements to achieving strong LLL performance. Thus, we propose Lifelong-MoE, an extensible MoE (Mixture-of-Experts) architecture that dynamically adds model capacity via adding experts with regularized pretraining. Our results show that by only introducing a limited number of extra experts while keeping the computation cost constant, our model can steadily adapt to data distribution shifts while preserving the previous knowledge. Compared to existing lifelong learning approaches, Lifelong-MoE achieves better few-shot performance on 19 downstream NLP tasks.

PaLM 2 Technical Report

We introduce PaLM 2, a new state-of-the-art language model that has better multilingual and reasoning capabilities and is more compute-efficient than its predecessor PaLM. PaLM 2 is a Transformer-based model trained using a mixture of objectives. Through extensive evaluations on English and multilingual language, and reasoning tasks, we demonstrate that PaLM 2 has significantly improved quality on downstream tasks across different model sizes, while simultaneously exhibiting faster and more efficient inference compared to PaLM. This improved efficiency enables broader deployment while also allowing the model to respond faster, for a more natural pace of interaction. PaLM 2 demonstrates robust reasoning capabilities exemplified by large improvements over PaLM on BIG-Bench and other reasoning tasks. PaLM 2 exhibits stable performance on a suite of responsible AI evaluations, and enables inference-time control over toxicity without additional overhead or impact on other capabilities. Overall, PaLM 2 achieves state-of-the-art performance across a diverse set of tasks and capabilities. When discussing the PaLM 2 family, it is important to distinguish between pre-trained models (of various sizes), fine-tuned variants of these models, and the user-facing products that use these models. In particular, user-facing products typically include additional pre- and post-processing steps. Additionally, the underlying models may evolve over time. Therefore, one should not expect the performance of user-facing products to exactly match the results reported in this report.

MaMMUT: A Simple Architecture for Joint Learning for MultiModal Tasks

The development of language models have moved from encoder-decoder to decoder-only designs. In addition, the common knowledge has it that the two most popular multimodal tasks, the generative and contrastive tasks, tend to conflict with one another, are hard to accommodate in one architecture, and further need complex adaptations for downstream tasks. We propose a novel paradigm of training with a decoder-only model for multimodal tasks, which is surprisingly effective in jointly learning of these disparate vision-language tasks. This is done with a simple model, called MaMMUT. It consists of a single vision encoder and a text decoder, and is able to accommodate contrastive and generative learning by a novel two-pass approach on the text decoder. We demonstrate that joint learning of these diverse objectives is simple, effective, and maximizes the weight-sharing of the model across these tasks. Furthermore, the same architecture enables straightforward extensions to open-vocabulary object detection and video-language tasks. The model tackles a diverse range of tasks, while being modest in capacity. Our model achieves the state of the art on image-text and text-image retrieval, video question answering and open-vocabulary detection tasks, outperforming much larger and more extensively trained foundational models. It shows very competitive results on VQA and Video Captioning, especially considering its capacity. Ablations confirm the flexibility and advantages of our approach.

AlpaServe: Statistical Multiplexing with Model Parallelism for Deep Learning Serving

Model parallelism is conventionally viewed as a method to scale a single large deep learning model beyond the memory limits of a single device. In this paper, we demonstrate that model parallelism can be additionally used for the statistical multiplexing of multiple devices when serving multiple models, even when a single model can fit into a single device. Our work reveals a fundamental trade-off between the overhead introduced by model parallelism and the opportunity to exploit statistical multiplexing to reduce serving latency in the presence of bursty workloads. We explore the new trade-off space and present a novel serving system, AlpaServe, that determines an efficient strategy for placing and parallelizing collections of large deep learning models across a distributed cluster. Evaluation results on production workloads show that AlpaServe can process requests at up to 10x higher rates or 6x more burstiness while staying within latency constraints for more than 99% of requests.

Massively Multilingual Shallow Fusion with Large Language Models

While large language models (LLM) have made impressive progress in natural language processing, it remains unclear how to utilize them in improving automatic speech recognition (ASR). In this work, we propose to train a single multilingual language model (LM) for shallow fusion in multiple languages. We push the limits of the multilingual LM to cover up to 84 languages by scaling up using a mixture-of-experts LLM, i.e., generalist language model (GLaM). When the number of experts increases, GLaM dynamically selects only two at each decoding step to keep the inference computation roughly constant. We then apply GLaM to a multilingual shallow fusion task based on a state-of-the-art end-to-end model. Compared to a dense LM of similar computation during inference, GLaM reduces the WER of an English long-tail test set by 4.4% relative. In a multilingual shallow fusion task, GLaM improves 41 out of 50 languages with an average relative WER reduction of 3.85%, and a maximum reduction of 10%. Compared to the baseline model, GLaM achieves an average WER reduction of 5.53% over 43 languages.

3D-EPI Blip-Up/Down Acquisition with CAIPI and Joint Hankel Structured Low-Rank Reconstruction for Rapid Distortion-Free High-Resolution T2* Mapping

Purpose: This work aims to develop a novel distortion-free 3D-EPI acquisition and image reconstruction technique for fast and robust, high-resolution, whole-brain imaging as well as quantitative T2* mapping. Methods: 3D-Blip-Up and -Down Acquisition (3D-BUDA) sequence is designed for both single- and multi-echo 3D GRE-EPI imaging using multiple shots with blip-up and -down readouts to encode B0 field map information. Complementary k-space coverage is achieved using controlled aliasing in parallel imaging (CAIPI) sampling across the shots. For image reconstruction, an iterative hard-thresholding algorithm is employed to minimize the cost function that combines field map information informed parallel imaging with the structured low-rank constraint for multi-shot 3D-BUDA data. Extending 3D-BUDA to multi-echo imaging permits T2* mapping. For this, we propose constructing a joint Hankel matrix along both echo and shot dimensions to improve the reconstruction. Results: Experimental results on in vivo multi-echo data demonstrate that, by performing joint reconstruction along with both echo and shot dimensions, reconstruction accuracy is improved compared to standard 3D-BUDA reconstruction. CAIPI sampling is further shown to enhance the image quality. For T2* mapping, T2* values from 3D-Joint-CAIPI-BUDA and reference multi-echo GRE are within limits of agreement as quantified by Bland-Altman analysis. Conclusions: The proposed technique enables rapid 3D distortion-free high-resolution imaging and T2* mapping. Specifically, 3D-BUDA enables 1-mm isotropic whole-brain imaging in 22 s at 3 T and 9 s on a 7 T scanner. The combination of multi-echo 3D-BUDA with CAIPI acquisition and joint reconstruction enables distortion-free whole-brain T2* mapping in 47 s at 1.1x1.1x1.0 mm3 resolution.

N-Grammer: Augmenting Transformers with latent n-grams

Transformer models have recently emerged as one of the foundational models in natural language processing, and as a byproduct, there is significant recent interest and investment in scaling these models. However, the training and inference costs of these large Transformer language models are prohibitive, thus necessitating more research in identifying more efficient variants. In this work, we propose a simple yet effective modification to the Transformer architecture inspired by the literature in statistical language modeling, by augmenting the model with n-grams that are constructed from a discrete latent representation of the text sequence. We evaluate our model, the N-Grammer on language modeling on the C4 data-set as well as text classification on the SuperGLUE data-set, and find that it outperforms several strong baselines such as the Transformer and the Primer. We open-source our model for reproducibility purposes in Jax.

Building Machine Translation Systems for the Next Thousand Languages

In this paper we share findings from our effort to build practical machine translation (MT) systems capable of translating across over one thousand languages. We describe results in three research domains: (i) Building clean, web-mined datasets for 1500+ languages by leveraging semi-supervised pre-training for language identification and developing data-driven filtering techniques; (ii) Developing practical MT models for under-served languages by leveraging massively multilingual models trained with supervised parallel data for over 100 high-resource languages and monolingual datasets for an additional 1000+ languages; and (iii) Studying the limitations of evaluation metrics for these languages and conducting qualitative analysis of the outputs from our MT models, highlighting several frequent error modes of these types of models. We hope that our work provides useful insights to practitioners working towards building MT systems for currently understudied languages, and highlights research directions that can complement the weaknesses of massively multilingual models in data-sparse settings.

Mixture-of-Experts with Expert Choice Routing

Sparsely-activated Mixture-of-experts (MoE) models allow the number of parameters to greatly increase while keeping the amount of computation for a given token or a given sample unchanged. However, a poor expert routing strategy (e.g. one resulting in load imbalance) can cause certain experts to be under-trained, leading to an expert being under or over-specialized. Prior work allocates a fixed number of experts to each token using a top-k function regardless of the relative importance of different tokens. To address this, we propose a heterogeneous mixture-of-experts employing an expert choice method. Instead of letting tokens select the top-k experts, we have experts selecting the top-k tokens. As a result, each token can be routed to a variable number of experts and each expert can have a fixed bucket size. We systematically study pre-training speedups using the same computational resources of the Switch Transformer top-1 and GShard top-2 gating of prior work and find that our method improves training convergence time by more than 2x. For the same computational cost, our method demonstrates higher performance in fine-tuning 11 selected tasks in the GLUE and SuperGLUE benchmarks. For a smaller activation cost, our method outperforms the T5 dense model in 7 out of the 11 tasks.

LaMDA: Language Models for Dialog Applications

We present LaMDA: Language Models for Dialog Applications. LaMDA is a family of Transformer-based neural language models specialized for dialog, which have up to 137B parameters and are pre-trained on 1.56T words of public dialog data and web text. While model scaling alone can improve quality, it shows less improvements on safety and factual grounding. We demonstrate that fine-tuning with annotated data and enabling the model to consult external knowledge sources can lead to significant improvements towards the two key challenges of safety and factual grounding. The first challenge, safety, involves ensuring that the model's responses are consistent with a set of human values, such as preventing harmful suggestions and unfair bias. We quantify safety using a metric based on an illustrative set of human values, and we find that filtering candidate responses using a LaMDA classifier fine-tuned with a small amount of crowdworker-annotated data offers a promising approach to improving model safety. The second challenge, factual grounding, involves enabling the model to consult external knowledge sources, such as an information retrieval system, a language translator, and a calculator. We quantify factuality using a groundedness metric, and we find that our approach enables the model to generate responses grounded in known sources, rather than responses that merely sound plausible. Finally, we explore the use of LaMDA in the domains of education and content recommendations, and analyze their helpfulness and role consistency.