arXiv subjects

Philipp Dufter

Publications and source records attributed to Philipp Dufter.

At least 19 recordsLinked to original sources

arXiv2025

MANZANO: A Simple and Scalable Unified Multimodal Model with a Hybrid Vision Tokenizer

Unified multimodal Large Language Models (LLMs) that can both understand and generate visual content hold immense potential. However, existing open-source models often suffer from a performance trade-off between these capabilities. We present Manzano, a simple and scalable unified framework that substantially reduces this tension by coupling a hybrid image tokenizer with a well-curated training recipe. A single shared vision encoder feeds two lightweight adapters that produce continuous embeddings for image-to-text understanding and discrete tokens for text-to-image generation within a common semantic space. A unified autoregressive LLM predicts high-level semantics in the form of text and image tokens, with an auxiliary diffusion decoder subsequently translating the image tokens into pixels. The architecture, together with a unified training recipe over understanding and generation data, enables scalable joint learning of both capabilities. Manzano achieves state-of-the-art results among unified models, and is competitive with specialist models, particularly on text-rich evaluation. Our studies show minimal task conflicts and consistent gains from scaling model size, validating our design choice of a hybrid tokenizer.

cs.CV↗

arXiv2025

AXLearn: Modular, Hardware-Agnostic Large Model Training

AXLearn is a production system which facilitates scalable and high-performance training of large deep learning models. Compared to other state-of-art deep learning systems, AXLearn has a unique focus on modularity and support for hardware-agnostic training. AXLearn's internal interfaces between software components follow strict encapsulation, allowing different components to be assembled to facilitate rapid model development and experimentation on different hardware infrastructure. AXLearn maintains constant complexity as we scale the components in the system, compared to linear or quadratic complexity in state-of-the-art training systems. This allows integrating features such as Rotary Position Embeddings (RoPE) into AXLearn across hundred of modules with just 10 lines of code, compared to hundreds as required in other systems. At the same time, AXLearn maintains equivalent performance compared to state-of-the-art training systems. Finally, we share our experience in the development and operation of AXLearn at Apple.

cs.LG↗

arXiv2024

Multimodal Autoregressive Pre-training of Large Vision Encoders

We introduce a novel method for pre-training of large-scale vision encoders. Building on recent advancements in autoregressive pre-training of vision models, we extend this framework to a multimodal setting, i.e., images and text. In this paper, we present AIMV2, a family of generalist vision encoders characterized by a straightforward pre-training process, scalability, and remarkable performance across a range of downstream tasks. This is achieved by pairing the vision encoder with a multimodal decoder that autoregressively generates raw image patches and text tokens. Our encoders excel not only in multimodal evaluations but also in vision benchmarks such as localization, grounding, and classification. Notably, our AIMV2-3B encoder achieves 89.5% accuracy on ImageNet-1k with a frozen trunk. Furthermore, AIMV2 consistently outperforms state-of-the-art contrastive models (e.g., CLIP, SigLIP) in multimodal image understanding across diverse settings.

cs.CV↗

arXiv2024

MM1.5: Methods, Analysis & Insights from Multimodal LLM Fine-tuning

We present MM1.5, a new family of multimodal large language models (MLLMs) designed to enhance capabilities in text-rich image understanding, visual referring and grounding, and multi-image reasoning. Building upon the MM1 architecture, MM1.5 adopts a data-centric approach to model training, systematically exploring the impact of diverse data mixtures across the entire model training lifecycle. This includes high-quality OCR data and synthetic captions for continual pre-training, as well as an optimized visual instruction-tuning data mixture for supervised fine-tuning. Our models range from 1B to 30B parameters, encompassing both dense and mixture-of-experts (MoE) variants, and demonstrate that careful data curation and training strategies can yield strong performance even at small scales (1B and 3B). Additionally, we introduce two specialized variants: MM1.5-Video, designed for video understanding, and MM1.5-UI, tailored for mobile UI understanding. Through extensive empirical studies and ablations, we provide detailed insights into the training processes and decisions that inform our final designs, offering valuable guidance for future research in MLLM development.

cs.CV↗

arXiv2024

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.

cs.CV↗

arXiv2024

Ferret-v2: An Improved Baseline for Referring and Grounding with Large Language Models

While Ferret seamlessly integrates regional understanding into the Large Language Model (LLM) to facilitate its referring and grounding capability, it poses certain limitations: constrained by the pre-trained fixed visual encoder and failed to perform well on broader tasks. In this work, we unveil Ferret-v2, a significant upgrade to Ferret, with three key designs. (1) Any resolution grounding and referring: A flexible approach that effortlessly handles higher image resolution, improving the model's ability to process and understand images in greater detail. (2) Multi-granularity visual encoding: By integrating the additional DINOv2 encoder, the model learns better and diverse underlying contexts for global and fine-grained visual information. (3) A three-stage training paradigm: Besides image-caption alignment, an additional stage is proposed for high-resolution dense alignment before the final instruction tuning. Experiments show that Ferret-v2 provides substantial improvements over Ferret and other state-of-the-art methods, thanks to its high-resolution scaling and fine-grained visual processing.

cs.CV↗

arXiv2022

Towards a Broad Coverage Named Entity Resource: A Data-Efficient Approach for Many Diverse Languages

Parallel corpora are ideal for extracting a multilingual named entity (MNE) resource, i.e., a dataset of names translated into multiple languages. Prior work on extracting MNE datasets from parallel corpora required resources such as large monolingual corpora or word aligners that are unavailable or perform poorly for underresourced languages. We present CLC-BN, a new method for creating an MNE resource, and apply it to the Parallel Bible Corpus, a corpus of more than 1000 languages. CLC-BN learns a neural transliteration model from parallel-corpus statistics, without requiring any other bilingual resources, word aligners, or seed data. Experimental results show that CLC-BN clearly outperforms prior work. We release an MNE resource for 1340 languages and demonstrate its effectiveness in two downstream tasks: knowledge graph augmentation and bilingual lexicon induction.

cs.CL↗

arXiv2021

BERT Cannot Align Characters

In previous work, it has been shown that BERT can adequately align cross-lingual sentences on the word level. Here we investigate whether BERT can also operate as a char-level aligner. The languages examined are English, Fake-English, German and Greek. We show that the closer two languages are, the better BERT can align them on the character level. BERT indeed works well in English to Fake-English alignment, but this does not generalize to natural languages to the same extent. Nevertheless, the proximity of two languages does seem to be a factor. English is more related to German than to Greek and this is reflected in how well BERT aligns them; English to German is better than English to Greek. We examine multiple setups and show that the similarity matrices for natural languages show weaker relations the further apart two languages are.

cs.CL↗

arXiv2021

Locating Language-Specific Information in Contextualized Embeddings

Multilingual pretrained language models (MPLMs) exhibit multilinguality and are well suited for transfer across languages. Most MPLMs are trained in an unsupervised fashion and the relationship between their objective and multilinguality is unclear. More specifically, the question whether MPLM representations are language-agnostic or they simply interleave well with learned task prediction heads arises. In this work, we locate language-specific information in MPLMs and identify its dimensionality and the layers where this information occurs. We show that language-specific information is scattered across many dimensions, which can be projected into a linear subspace. Our study contributes to a better understanding of MPLM representations, going beyond treating them as unanalyzable blobs of information.

cs.CL↗

arXiv2021

Wine is Not v i n. -- On the Compatibility of Tokenizations Across Languages

The size of the vocabulary is a central design choice in large pretrained language models, with respect to both performance and memory requirements. Typically, subword tokenization algorithms such as byte pair encoding and WordPiece are used. In this work, we investigate the compatibility of tokenizations for multilingual static and contextualized embedding spaces and propose a measure that reflects the compatibility of tokenizations across languages. Our goal is to prevent incompatible tokenizations, e.g., "wine" (word-level) in English vs.\ "v i n" (character-level) in French, which make it hard to learn good multilingual semantic representations. We show that our compatibility measure allows the system designer to create vocabularies across languages that are compatible -- a desideratum that so far has been neglected in multilingual models.

cs.CL↗

arXiv2021

Graph Algorithms for Multiparallel Word Alignment

With the advent of end-to-end deep learning approaches in machine translation, interest in word alignments initially decreased; however, they have again become a focus of research more recently. Alignments are useful for typological research, transferring formatting like markup to translated texts, and can be used in the decoding of machine translation systems. At the same time, massively multilingual processing is becoming an important NLP scenario, and pretrained language and machine translation models that are truly multilingual are proposed. However, most alignment algorithms rely on bitexts only and do not leverage the fact that many parallel corpora are multiparallel. In this work, we exploit the multiparallelity of corpora by representing an initial set of bilingual alignments as a graph and then predicting additional edges in the graph. We present two graph algorithms for edge prediction: one inspired by recommender systems and one based on network link prediction. Our experimental results show absolute improvements in $F_1$ of up to 28% over the baseline bilingual word aligner in different datasets.

cs.CL↗

arXiv2021

Position Information in Transformers: An Overview

Transformers are arguably the main workhorse in recent Natural Language Processing research. By definition a Transformer is invariant with respect to reordering of the input. However, language is inherently sequential and word order is essential to the semantics and syntax of an utterance. In this article, we provide an overview and theoretical comparison of existing methods to incorporate position information into Transformer models. The objectives of this survey are to (1) showcase that position information in Transformer is a vibrant and extensive research area; (2) enable the reader to compare existing methods by providing a unified notation and systematization of different approaches along important model dimensions; (3) indicate what characteristics of an application should be taken into account when selecting a position encoding; (4) provide stimuli for future research.

cs.CL↗

arXiv2021

ParCourE: A Parallel Corpus Explorer for a Massively Multilingual Corpus

With more than 7000 languages worldwide, multilingual natural language processing (NLP) is essential both from an academic and commercial perspective. Researching typological properties of languages is fundamental for progress in multilingual NLP. Examples include assessing language similarity for effective transfer learning, injecting inductive biases into machine learning models or creating resources such as dictionaries and inflection tables. We provide ParCourE, an online tool that allows to browse a word-aligned parallel corpus, covering 1334 languages. We give evidence that this is useful for typological research. ParCourE can be set up for any parallel corpus and can thus be used for typological research on other corpora as well as for exploring their quality and properties.

cs.CL↗

arXiv2021

Subword Sampling for Low Resource Word Alignment

Annotation projection is an important area in NLP that can greatly contribute to creating language resources for low-resource languages. Word alignment plays a key role in this setting. However, most of the existing word alignment methods are designed for a high resource setting in machine translation where millions of parallel sentences are available. This amount reduces to a few thousands of sentences when dealing with low-resource languages failing the existing established IBM models. In this paper, we propose subword sampling-based alignment of text units. This method's hypothesis is that the aggregation of different granularities of text for certain language pairs can help word-level alignment. For certain languages for which gold-standard alignments exist, we propose an iterative Bayesian optimization framework to optimize selecting possible subwords from the space of possible subword representations of the source and target sentences. We show that the subword sampling method consistently outperforms word-level alignment on six language pairs: English-German, English-French, English-Romanian, English-Persian, English-Hindi, and English-Inuktitut. In addition, we show that the hyperparameters learned for certain language pairs can be applied to other languages at no supervision and consistently improve the alignment results. We observe that using $5K$ parallel sentences together with our proposed subword sampling approach, we obtain similar F1 scores to the use of $100K$'s of parallel sentences in existing word-level fast-align/eflomal alignment methods.

cs.CL↗

arXiv2021

Modeling Graph Structure via Relative Position for Text Generation from Knowledge Graphs

We present Graformer, a novel Transformer-based encoder-decoder architecture for graph-to-text generation. With our novel graph self-attention, the encoding of a node relies on all nodes in the input graph - not only direct neighbors - facilitating the detection of global patterns. We represent the relation between two nodes as the length of the shortest path between them. Graformer learns to weight these node-node relations differently for different attention heads, thus virtually learning differently connected views of the input graph. We evaluate Graformer on two popular graph-to-text generation benchmarks, AGENDA and WebNLG, where it achieves strong performance while using many fewer parameters than other approaches.

cs.CL↗

arXiv2021

SimAlign: High Quality Word Alignments without Parallel Training Data using Static and Contextualized Embeddings

Word alignments are useful for tasks like statistical and neural machine translation (NMT) and cross-lingual annotation projection. Statistical word aligners perform well, as do methods that extract alignments jointly with translations in NMT. However, most approaches require parallel training data, and quality decreases as less training data is available. We propose word alignment methods that require no parallel data. The key idea is to leverage multilingual word embeddings, both static and contextualized, for word alignment. Our multilingual embeddings are created from monolingual data only without relying on any parallel data or dictionaries. We find that alignments created from embeddings are superior for four and comparable for two language pairs compared to those produced by traditional statistical aligners, even with abundant parallel data; e.g., contextualized embeddings achieve a word alignment F1 for English-German that is 5 percentage points higher than eflomal, a high-quality statistical aligner, trained on 100k parallel sentences.

cs.CL↗

arXiv2021

Static Embeddings as Efficient Knowledge Bases?

Recent research investigates factual knowledge stored in large pretrained language models (PLMs). Instead of structural knowledge base (KB) queries, masked sentences such as "Paris is the capital of [MASK]" are used as probes. The good performance on this analysis task has been interpreted as PLMs becoming potential repositories of factual knowledge. In experiments across ten linguistically diverse languages, we study knowledge contained in static embeddings. We show that, when restricting the output space to a candidate set, simple nearest neighbor matching using static embeddings performs better than PLMs. E.g., static embeddings perform 1.6% points better than BERT while just using 0.3% of energy for training. One important factor in their good comparative performance is that static embeddings are standardly learned for a large vocabulary. In contrast, BERT exploits its more sophisticated, but expensive ability to compose meaningful representations from a much smaller subword vocabulary.

cs.CL↗

arXiv2021

Identifying Necessary Elements for BERT's Multilinguality

It has been shown that multilingual BERT (mBERT) yields high quality multilingual representations and enables effective zero-shot transfer. This is surprising given that mBERT does not use any crosslingual signal during training. While recent literature has studied this phenomenon, the reasons for the multilinguality are still somewhat obscure. We aim to identify architectural properties of BERT and linguistic properties of languages that are necessary for BERT to become multilingual. To allow for fast experimentation we propose an efficient setup with small BERT models trained on a mix of synthetic and natural data. Overall, we identify four architectural and two linguistic elements that influence multilinguality. Based on our insights, we experiment with a multilingual pretraining setup that modifies the masking strategy using VecMap, i.e., unsupervised embedding alignment. Experiments on XNLI with three languages indicate that our findings transfer from our small setup to larger scale settings.

cs.CL↗