Harder Tasks Need More Experts: Dynamic Routing in MoE Models

In this paper, we introduce a novel dynamic expert selection framework for Mixture of Experts (MoE) models, aiming to enhance computational efficiency and model performance by adjusting the number of activated experts based on input difficulty. Unlike traditional MoE approaches that rely on fixed Top-K routing, which activates a predetermined number of experts regardless of the input's complexity, our method dynamically selects experts based on the confidence level in expert selection for each input. This allows for a more efficient utilization of computational resources, activating more experts for complex tasks requiring advanced reasoning and fewer for simpler tasks. Through extensive evaluations, our dynamic routing method demonstrates substantial improvements over conventional Top-2 routing across various benchmarks, achieving an average improvement of 0.7% with less than 90% activated parameters. Further analysis shows our model dispatches more experts to tasks requiring complex reasoning skills, like BBH, confirming its ability to dynamically allocate computational resources in alignment with the input's complexity. Our findings also highlight a variation in the number of experts needed across different layers of the transformer model, offering insights into the potential for designing heterogeneous MoE frameworks. The code and models are available at https://github.com/ZhenweiAn/Dynamic_MoE.

BUS:Efficient and Effective Vision-language Pre-training with Bottom-Up Patch Summarization

Vision Transformer (ViT) based Vision-Language Pre-training (VLP) models have demonstrated impressive performance in various tasks. However, the lengthy visual token sequences fed into ViT can lead to training inefficiency and ineffectiveness. Existing efforts address the challenge by either bottom-level patch extraction in the ViT backbone or top-level patch abstraction outside, not balancing training efficiency and effectiveness well. Inspired by text summarization in natural language processing, we propose a Bottom-Up Patch Summarization approach named BUS, coordinating bottom-level extraction and top-level abstraction to learn a concise summary of lengthy visual token sequences efficiently. Specifically, We incorporate a Text-Semantics-Aware Patch Selector (TSPS) into the ViT backbone to perform a coarse-grained visual token extraction and then attach a flexible Transformer-based Patch Abstraction Decoder (PAD) upon the backbone for top-level visual abstraction. This bottom-up collaboration enables our BUS to yield high training efficiency while maintaining or even improving effectiveness. We evaluate our approach on various visual-language understanding and generation tasks and show competitive downstream task performance while boosting the training efficiency by 50\%. Additionally, our model achieves state-of-the-art performance on many downstream tasks by increasing input image resolution without increasing computational costs over baselines.

DiffDTM: A conditional structure-free framework for bioactive molecules generation targeted for dual proteins

Advances in deep generative models shed light on de novo molecule generation with desired properties. However, molecule generation targeted for dual protein targets still faces formidable challenges including protein 3D structure data requisition for model training, auto-regressive sampling, and model generalization for unseen targets. Here, we proposed DiffDTM, a novel conditional structure-free deep generative model based on a diffusion model for dual targets based molecule generation to address the above issues. Specifically, DiffDTM receives protein sequences and molecular graphs as inputs instead of protein and molecular conformations and incorporates an information fusion module to achieve conditional generation in a one-shot manner. We have conducted comprehensive multi-view experiments to demonstrate that DiffDTM can generate drug-like, synthesis-accessible, novel, and high-binding affinity molecules targeting specific dual proteins, outperforming the state-of-the-art (SOTA) models in terms of multiple evaluation metrics. Furthermore, we utilized DiffDTM to generate molecules towards dopamine receptor D2 and 5-hydroxytryptamine receptor 1A as new antipsychotics. The experimental results indicate that DiffDTM can be easily plugged into unseen dual targets to generate bioactive molecules, addressing the issues of requiring insufficient active molecule data for training as well as the need to retrain when encountering new targets.

Towards Adaptive Prefix Tuning for Parameter-Efficient Language Model Fine-tuning

Fine-tuning large pre-trained language models on various downstream tasks with whole parameters is prohibitively expensive. Hence, Parameter-efficient fine-tuning has attracted attention that only optimizes a few task-specific parameters with the frozen pre-trained model. In this work, we focus on prefix tuning, which only optimizes continuous prefix vectors (i.e. pseudo tokens) inserted into Transformer layers. Based on the observation that the learned syntax and semantics representation varies a lot at different layers, we argue that the adaptive prefix will be further tailored to each layer than the fixed one, enabling the fine-tuning more effective and efficient. Thus, we propose Adaptive Prefix Tuning (APT) to adjust the prefix in terms of both fine-grained token level and coarse-grained layer level with a gate mechanism. Experiments on the SuperGLUE and NER datasets show the effectiveness of APT. In addition, taking the gate as a probing, we validate the efficiency and effectiveness of the variable prefix.

VECO 2.0: Cross-lingual Language Model Pre-training with Multi-granularity Contrastive Learning

Recent studies have demonstrated the potential of cross-lingual transferability by training a unified Transformer encoder for multiple languages. In addition to involving the masked language model objective, existing cross-lingual pre-training works leverage sentence-level contrastive learning or plugs in extra cross-attention module to complement the insufficient capabilities of cross-lingual alignment. Nonetheless, synonym pairs residing in bilingual corpus are not exploited and aligned, which is more crucial than sentence interdependence establishment for token-level tasks. In this work, we propose a cross-lingual pre-trained model VECO~2.0 based on contrastive learning with multi-granularity alignments. Specifically, the sequence-to-sequence alignment is induced to maximize the similarity of the parallel pairs and minimize the non-parallel pairs. Then, token-to-token alignment is integrated to bridge the gap between synonymous tokens excavated via the thesaurus dictionary from the other unpaired tokens in a bilingual instance. Experiments show the effectiveness of the proposed strategy for cross-lingual model pre-training on the XTREME benchmark.

RRHF: Rank Responses to Align Language Models with Human Feedback without tears

Reinforcement Learning from Human Feedback (RLHF) facilitates the alignment of large language models with human preferences, significantly enhancing the quality of interactions between humans and these models. InstructGPT implements RLHF through several stages, including Supervised Fine-Tuning (SFT), reward model training, and Proximal Policy Optimization (PPO). PPO, however, is sensitive to hyperparameters and requires a minimum of four models in its standard implementation, which makes it hard to train. In contrast, we propose a novel learning paradigm called RRHF, which scores responses generated by different sampling policies and learns to align them with human preferences through ranking loss. RRHF can efficiently align language model output probabilities with human preferences as robust as fine-tuning and it only needs 1 to 2 models during tuning. In addition, RRHF can be considered an extension of SFT and reward models while being simpler than PPO in terms of coding, model counts, and hyperparameters. The entire alignment process can be accomplished within a single RRHF training session. We evaluate RRHF using LLaMA and Alpaca on Helpful and Harmless data, demonstrating performance comparable to PPO.

Revisiting Automatic Question Summarization Evaluation in the Biomedical Domain

Automatic evaluation metrics have been facilitating the rapid development of automatic summarization methods by providing instant and fair assessments of the quality of summaries. Most metrics have been developed for the general domain, especially news and meeting notes, or other language-generation tasks. However, these metrics are applied to evaluate summarization systems in different domains, such as biomedical question summarization. To better understand whether commonly used evaluation metrics are capable of evaluating automatic summarization in the biomedical domain, we conduct human evaluations of summarization quality from four different aspects of a biomedical question summarization task. Based on human judgments, we identify different noteworthy features for current automatic metrics and summarization systems as well. We also release a dataset of our human annotations to aid the research of summarization evaluation metrics in the biomedical domain.

RAMM: Retrieval-augmented Biomedical Visual Question Answering with Multi-modal Pre-training

Vision-and-language multi-modal pretraining and fine-tuning have shown great success in visual question answering (VQA). Compared to general domain VQA, the performance of biomedical VQA suffers from limited data. In this paper, we propose a retrieval-augmented pretrain-and-finetune paradigm named RAMM for biomedical VQA to overcome the data limitation issue. Specifically, we collect a new biomedical dataset named PMCPM which offers patient-based image-text pairs containing diverse patient situations from PubMed. Then, we pretrain the biomedical multi-modal model to learn visual and textual representation for image-text pairs and align these representations with image-text contrastive objective (ITC). Finally, we propose a retrieval-augmented method to better use the limited data. We propose to retrieve similar image-text pairs based on ITC from pretraining datasets and introduce a novel retrieval-attention module to fuse the representation of the image and the question with the retrieved images and texts. Experiments demonstrate that our retrieval-augmented pretrain-and-finetune paradigm obtains state-of-the-art performance on Med-VQA2019, Med-VQA2021, VQARAD, and SLAKE datasets. Further analysis shows that the proposed RAMM and PMCPM can enhance biomedical VQA performance compared with previous resources and methods. We will open-source our dataset, codes, and pretrained model.

Molecular Geometry-aware Transformer for accurate 3D Atomic System modeling

Molecular dynamic simulations are important in computational physics, chemistry, material, and biology. Machine learning-based methods have shown strong abilities in predicting molecular energy and properties and are much faster than DFT calculations. Molecular energy is at least related to atoms, bonds, bond angles, torsion angles, and nonbonding atom pairs. Previous Transformer models only use atoms as inputs which lack explicit modeling of the aforementioned factors. To alleviate this limitation, we propose Moleformer, a novel Transformer architecture that takes nodes (atoms) and edges (bonds and nonbonding atom pairs) as inputs and models the interactions among them using rotational and translational invariant geometry-aware spatial encoding. Proposed spatial encoding calculates relative position information including distances and angles among nodes and edges. We benchmark Moleformer on OC20 and QM9 datasets, and our model achieves state-of-the-art on the initial state to relaxed energy prediction of OC20 and is very competitive in QM9 on predicting quantum chemical properties compared to other Transformer and Graph Neural Network (GNN) methods which proves the effectiveness of the proposed geometry-aware spatial encoding in Moleformer.

mPLUG-2: A Modularized Multi-modal Foundation Model Across Text, Image and Video

Recent years have witnessed a big convergence of language, vision, and multi-modal pretraining. In this work, we present mPLUG-2, a new unified paradigm with modularized design for multi-modal pretraining, which can benefit from modality collaboration while addressing the problem of modality entanglement. In contrast to predominant paradigms of solely relying on sequence-to-sequence generation or encoder-based instance discrimination, mPLUG-2 introduces a multi-module composition network by sharing common universal modules for modality collaboration and disentangling different modality modules to deal with modality entanglement. It is flexible to select different modules for different understanding and generation tasks across all modalities including text, image, and video. Empirical study shows that mPLUG-2 achieves state-of-the-art or competitive results on a broad range of over 30 downstream tasks, spanning multi-modal tasks of image-text and video-text understanding and generation, and uni-modal tasks of text-only, image-only, and video-only understanding. Notably, mPLUG-2 shows new state-of-the-art results of 48.0 top-1 accuracy and 80.3 CIDEr on the challenging MSRVTT video QA and video caption tasks with a far smaller model size and data scale. It also demonstrates strong zero-shot transferability on vision-language and video-language tasks. Code and models will be released in https://github.com/alibaba/AliceMind.