Search

Efficiently modeling sequences with infinite context length has been a long-standing problem. Past works suffer from either the quadratic computation complexity or the limited extrapolation ability on length generalization. In this work, we present Samba, a simple hybrid architecture that layer-wise combines Mamba, a selective State Space Model (SSM), with Sliding Window Attention (SWA). Samba selectively compresses a given sequence into recurrent hidden states while still maintaining the ability to precisely recall memories with the attention mechanism. We scale Samba up to 3.8B parameters with 3.2T training tokens and show that Samba substantially outperforms the state-of-the-art models based on pure attention or SSMs on a wide range of benchmarks. When trained on 4K length sequences, Samba can be efficiently extrapolated to 256K context length with perfect memory recall and show improved token predictions up to 1M context length. As a linear-time sequence model, Samba enjoys a 3.73x higher throughput compared to Transformers with grouped-query attention when processing user prompts of 128K length, and 3.64x speedup when generating 64K tokens with unlimited streaming. A sample implementation of Samba is publicly available in https://github.com/microsoft/Samba.

We introduce phi-3-mini, a 3.8 billion parameter language model trained on 3.3 trillion tokens, whose overall performance, as measured by both academic benchmarks and internal testing, rivals that of models such as Mixtral 8x7B and GPT-3.5 (e.g., phi-3-mini achieves 69% on MMLU and 8.38 on MT-bench), despite being small enough to be deployed on a phone. Our training dataset is a scaled-up version of the one used for phi-2, composed of heavily filtered publicly available web data and synthetic data. The model is also further aligned for robustness, safety, and chat format. We also provide parameter-scaling results with a 7B, 14B models trained for 4.8T tokens, called phi-3-small, phi-3-medium, both significantly more capable than phi-3-mini (e.g., respectively 75%, 78% on MMLU, and 8.7, 8.9 on MT-bench). To enhance multilingual, multimodal, and long-context capabilities, we introduce three models in the phi-3.5 series: phi-3.5-mini, phi-3.5-MoE, and phi-3.5-Vision. The phi-3.5-MoE, a 16 x 3.8B MoE model with 6.6 billion active parameters, achieves superior performance in language reasoning, math, and code tasks compared to other open-source models of similar scale, such as Llama 3.1 and the Mixtral series, and on par with Gemini-1.5-Flash and GPT-4o-mini. Meanwhile, phi-3.5-Vision, a 4.2 billion parameter model derived from phi-3.5-mini, excels in reasoning tasks and is adept at handling both single-image and text prompts, as well as multi-image and text prompts., Comment: 24 pages

Existing reference-free turn-level evaluation metrics for chatbots inadequately capture the interaction between the user and the system. Consequently, they often correlate poorly with human evaluations. To address this issue, we propose a novel model-agnostic approach that leverages Conditional Pointwise Mutual Information (C-PMI) to measure the turn-level interaction between the system and the user based on a given evaluation dimension. Experimental results on the widely used FED dialogue evaluation dataset demonstrate that our approach significantly improves the correlation with human judgment compared with existing evaluation systems. By replacing the negative log-likelihood-based scorer with our proposed C-PMI scorer, we achieve a relative 62.6% higher Spearman correlation on average for the FED evaluation metric. Our code is publicly available at https://github.com/renll/C-PMI., Comment: Published at ACL2023 DialDoc Workshop; Updated Results

Recent hybrid models combining Linear State Space Models (SSMs) with self-attention mechanisms have demonstrated impressive results across a range of sequence modeling tasks. However, current approaches apply attention modules statically and uniformly to all elements in the input sequences, leading to sub-optimal quality-efficiency trade-offs. To address this limitation, we introduce Sparse Modular Activation (SMA), a general mechanism enabling neural networks to sparsely and dynamically activate sub-modules for sequence elements in a differentiable manner. Through allowing each element to skip non-activated sub-modules, SMA reduces computation and memory consumption of neural networks at both training and inference stages. To validate the effectiveness of SMA on sequence modeling, we design a novel neural architecture, SeqBoat, which employs SMA to sparsely activate a Gated Attention Unit (GAU) based on the state representations learned from an SSM. By constraining the GAU to only conduct local attention on the activated inputs, SeqBoat can achieve linear inference complexity with theoretically infinite attention span, and provide substantially better quality-efficiency trade-off than the chunking-based models. With experiments on a wide range of tasks, including long sequence modeling, speech classification and language modeling, SeqBoat brings new state-of-the-art results among hybrid models with linear complexity, and reveals the amount of attention needed for each task through the learned sparse activation patterns. Our code is publicly available at https://github.com/renll/SeqBoat., Comment: Accepted by NeurIPS 2023. Camera-ready Version

Modern large-scale Pre-trained Language Models (PLMs) have achieved tremendous success on a wide range of downstream tasks. However, most of the LM pre-training objectives only focus on text reconstruction, but have not sought to learn latent-level interpretable representations of sentences. In this paper, we manage to push the language models to obtain a deeper understanding of sentences by proposing a new pre-training objective, Sparse Latent Typing, which enables the model to sparsely extract sentence-level keywords with diverse latent types. Experimental results show that our model is able to learn interpretable latent type categories in a self-supervised manner without using any external knowledge. Besides, the language model pre-trained with such an objective also significantly improves Information Extraction related downstream tasks in both supervised and few-shot settings. Our code is publicly available at: https://github.com/renll/SparseLT., Comment: EMNLP 2022 (Oral)

Text-to-Graph extraction aims to automatically extract information graphs consisting of mentions and types from natural language texts. Existing approaches, such as table filling and pairwise scoring, have shown impressive performance on various information extraction tasks, but they are difficult to scale to datasets with longer input texts because of their second-order space/time complexities with respect to the input length. In this work, we propose a Hybrid Span Generator (HySPA) that invertibly maps the information graph to an alternating sequence of nodes and edge types, and directly generates such sequences via a hybrid span decoder which can decode both the spans and the types recurrently in linear time and space complexities. Extensive experiments on the ACE05 dataset show that our approach also significantly outperforms state-of-the-art on the joint entity and relation extraction task., Comment: Accepted by ACL 2021 Findings

Convolutional Neural Networks (CNN) has been widely applied in the realm of computer vision. However, given the fact that CNN models are translation invariant, they are not aware of the coordinate information of each pixel. Thus the generalization ability of CNN will be limited since the coordinate information is crucial for a model to learn affine transformations which directly operate on the coordinate of each pixel. In this project, we proposed a simple approach to incorporate the coordinate information to the CNN model through coordinate embedding. Our approach does not change the downstream model architecture and can be easily applied to the pre-trained models for the task like object detection. Our experiments on the German Traffic Sign Detection Benchmark show that our approach not only significantly improve the model performance but also have better robustness with respect to the affine transformation., Comment: 6 pages, 8 figures, Course Project for ECE271B

Natural gradient has been recently introduced to the field of boosting to enable the generic probabilistic predication capability. Natural gradient boosting shows promising performance improvements on small datasets due to better training dynamics, but it suffers from slow training speed overhead especially for large datasets. We present a replication study of NGBoost(Duan et al., 2019) training that carefully examines the impacts of key hyper-parameters under the circumstance of best-first decision tree learning. We find that with the regularization of leaf number clipping, the performance of NGBoost can be largely improved via a better choice of hyperparameters. Experiments show that our approach significantly beats the state-of-the-art performance on various kinds of datasets from the UCI Machine Learning Repository while still has up to 4.85x speed up compared with the original approach of NGBoost.

Existing approaches to dialogue state tracking rely on pre-defined ontologies consisting of a set of all possible slot types and values. Though such approaches exhibit promising performance on single-domain benchmarks, they suffer from computational complexity that increases proportionally to the number of pre-defined slots that need tracking. This issue becomes more severe when it comes to multi-domain dialogues which include larger numbers of slots. In this paper, we investigate how to approach DST using a generation framework without the pre-defined ontology list. Given each turn of user utterance and system response, we directly generate a sequence of belief states by applying a hierarchical encoder-decoder structure. In this way, the computational complexity of our model will be a constant regardless of the number of pre-defined slots. Experiments on both the multi-domain and the single domain dialogue state tracking dataset show that our model not only scales easily with the increasing number of pre-defined domains and slots but also reaches the state-of-the-art performance., Comment: The 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP 2019); Updated empirical results

Dialogue state tracking is the core part of a spoken dialogue system. It estimates the beliefs of possible user's goals at every dialogue turn. However, for most current approaches, it's difficult to scale to large dialogue domains. They have one or more of following limitations: (a) Some models don't work in the situation where slot values in ontology changes dynamically; (b) The number of model parameters is proportional to the number of slots; (c) Some models extract features based on hand-crafted lexicons. To tackle these challenges, we propose StateNet, a universal dialogue state tracker. It is independent of the number of values, shares parameters across all slots, and uses pre-trained word vectors instead of explicit semantic dictionaries. Our experiments on two datasets show that our approach not only overcomes the limitations, but also significantly outperforms the performance of state-of-the-art approaches., Comment: EMNLP 2018 Oral

We propose the Recurrent Soft Attention Model, which integrates the visual attention from the original image to a LSTM memory cell through a down-sample network. The model recurrently transmits visual attention to the memory cells for glimpse mask generation, which is a more natural way for attention integration and exploitation in general object detection and recognition problem. We test our model under the metric of the top-1 accuracy on the CIFAR-10 dataset. The experiment shows that our down-sample network and feedback mechanism plays an effective role among the whole network structure., Comment: 5 pages, 4 figures