LEGATO: Large-scale End-to-end Generalizable Approach to Typeset OMR

We propose Legato, a new end-to-end transformer model for optical music recognition (OMR). Legato is the first large-scale pretrained OMR model capable of recognizing full-page or multi-page typeset music scores and the first to generate documents in ABC notation, a concise, human-readable format for symbolic music. Bringing together a pretrained vision encoder with an ABC decoder trained on a dataset of more than 214K images, our model exhibits the strong ability to generalize across various typeset scores. We conduct experiments on a range of datasets and demonstrate that our model achieves state-of-the-art performance. Given the lack of a standardized evaluation for end-to-end OMR, we comprehensively compare our model against the previous state of the art using a diverse set of metrics.

Broken Tokens? Your Language Model can Secretly Handle Non-Canonical Tokenizations

Modern tokenizers employ deterministic algorithms to map text into a single "canonical" token sequence, yet the same string can be encoded as many non-canonical tokenizations using the tokenizer vocabulary. In this work, we investigate the robustness of LMs to text encoded with non-canonical tokenizations entirely unseen during training. Surprisingly, when evaluated across 20 benchmarks, we find that instruction-tuned models retain up to 93.4% of their original performance when given a randomly sampled tokenization, and 90.8% with character-level tokenization. We see that overall stronger models tend to be more robust, and robustness diminishes as the tokenization departs farther from the canonical form. Motivated by these results, we then identify settings where non-canonical tokenization schemes can *improve* performance, finding that character-level segmentation improves string manipulation and code understanding tasks by up to +14%, and right-aligned digit grouping enhances large-number arithmetic by +33%. Finally, we investigate the source of this robustness, finding that it arises in the instruction-tuning phase. We show that while both base and post-trained models grasp the semantics of non-canonical tokenizations (perceiving them as containing misspellings), base models try to mimic the imagined mistakes and degenerate into nonsensical output, while post-trained models are committed to fluent responses. Overall, our findings suggest that models are less tied to their tokenizer than previously believed, and demonstrate the promise of intervening on tokenization at inference time to boost performance.

Sampling from Your Language Model One Byte at a Time

Tokenization is used almost universally by modern language models, enabling efficient text representation using multi-byte or multi-character tokens. However, prior work has shown that tokenization can introduce distortion into the model's generations. For example, users are often advised not to end their prompts with a space because it prevents the model from including the space as part of the next token. This Prompt Boundary Problem (PBP) also arises in languages such as Chinese and in code generation, where tokens often do not line up with syntactic boundaries. Additionally mismatching tokenizers often hinder model composition and interoperability. For example, it is not possible to directly ensemble models with different tokenizers due to their mismatching vocabularies. To address these issues, we present an inference-time method to convert any autoregressive LM with a BPE tokenizer into a character-level or byte-level LM, without changing its generative distribution at the text level. Our method efficient solves the PBP and is also able to unify the vocabularies of language models with different tokenizers, allowing one to ensemble LMs with different tokenizers at inference time as well as transfer the post-training from one model to another using proxy-tuning. We demonstrate in experiments that the ensemble and proxy-tuned models outperform their constituents on downstream evals.

Infini-gram mini: Exact n-gram Search at the Internet Scale with FM-Index

Language models are trained mainly on massive text data from the Internet, and it becomes increasingly important to understand this data source. Exact-match search engines enable searching in large text corpora -- counting string appearances and retrieving the enclosing documents -- yet the high storage overhead hinders their application on Internet-scale data. We present Infini-gram mini, an efficient and scalable system that can make petabyte-level text corpora searchable. Based on the FM-index data structure (Ferragina and Manzini, 2000), which simultaneously indexes and compresses text, our system creates indexes with size only 44% of the corpus. Infini-gram mini greatly improves upon the best existing implementation of FM-index in terms of indexing speed (18$\times$) and memory use during both indexing (3.2$\times$ reduction) and querying (down to a negligible amount). We index 46TB of Internet text in 50 days with a single 128-core CPU node (or 19 hours if using 75 such nodes). We show one important use case of Infini-gram mini in a large-scale analysis of benchmark contamination. We find several core LM evaluation benchmarks to be heavily contaminated in Internet crawls (up to 40% in SQuAD), which could lead to overestimating the capabilities of language models if trained on such data. We host a benchmark contamination bulletin to share the contamination rate of many core and community-contributed benchmarks. We also release a web interface and an API endpoint to serve general search queries on Infini-gram mini indexes.

MMMG: a Comprehensive and Reliable Evaluation Suite for Multitask Multimodal Generation

Automatically evaluating multimodal generation presents a significant challenge, as automated metrics often struggle to align reliably with human evaluation, especially for complex tasks that involve multiple modalities. To address this, we present MMMG, a comprehensive and human-aligned benchmark for multimodal generation across 4 modality combinations (image, audio, interleaved text and image, interleaved text and audio), with a focus on tasks that present significant challenges for generation models, while still enabling reliable automatic evaluation through a combination of models and programs. MMMG encompasses 49 tasks (including 29 newly developed ones), each with a carefully designed evaluation pipeline, and 937 instructions to systematically assess reasoning, controllability, and other key capabilities of multimodal generation models. Extensive validation demonstrates that MMMG is highly aligned with human evaluation, achieving an average agreement of 94.3%. Benchmarking results on 24 multimodal generation models reveal that even though the state-of-the-art model, GPT Image, achieves 78.3% accuracy for image generation, it falls short on multimodal reasoning and interleaved generation. Furthermore, results suggest considerable headroom for improvement in audio generation, highlighting an important direction for future research.

PointArena: Probing Multimodal Grounding Through Language-Guided Pointing

Pointing serves as a fundamental and intuitive mechanism for grounding language within visual contexts, with applications spanning robotics, assistive technologies, and interactive AI systems. While recent multimodal models have started to support pointing capabilities, existing benchmarks typically focus only on referential object localization tasks. We introduce PointArena, a comprehensive platform for evaluating multimodal pointing across diverse reasoning scenarios. PointArena comprises three components: (1) Point-Bench, a curated dataset containing approximately 1,000 pointing tasks across five reasoning categories; (2) Point-Battle, an interactive, web-based arena facilitating blind, pairwise model comparisons, which has already gathered over 4,500 anonymized votes; and (3) Point-Act, a real-world robotic manipulation system allowing users to directly evaluate multimodal model pointing capabilities in practical settings. We conducted extensive evaluations of both state-of-the-art open-source and proprietary multimodal models. Results indicate that Molmo-72B consistently outperforms other models, though proprietary models increasingly demonstrate comparable performance. Additionally, we find that supervised training specifically targeting pointing tasks significantly enhances model performance. Across our multi-stage evaluation pipeline, we also observe strong correlations, underscoring the critical role of precise pointing capabilities in enabling multimodal models to effectively bridge abstract reasoning with concrete, real-world actions. Project page: https://pointarena.github.io/

BLAB: Brutally Long Audio Bench

Developing large audio language models (LMs) capable of understanding diverse spoken interactions is essential for accommodating the multimodal nature of human communication and can increase the accessibility of language technologies across different user populations. Recent work on audio LMs has primarily evaluated their performance on short audio segments, typically under 30 seconds, with limited exploration of long-form conversational speech segments that more closely reflect natural user interactions with these models. We introduce Brutally Long Audio Bench (BLAB), a challenging long-form audio benchmark that evaluates audio LMs on localization, duration estimation, emotion, and counting tasks using audio segments averaging 51 minutes in length. BLAB consists of 833+ hours of diverse, full-length audio clips, each paired with human-annotated, text-based natural language questions and answers. Our audio data were collected from permissively licensed sources and underwent a human-assisted filtering process to ensure task compliance. We evaluate six open-source and proprietary audio LMs on BLAB and find that all of them, including advanced models such as Gemini 2.0 Pro and GPT-4o, struggle with the tasks in BLAB. Our comprehensive analysis reveals key insights into the trade-offs between task difficulty and audio duration. In general, we find that audio LMs struggle with long-form speech, with performance declining as duration increases. They perform poorly on localization, temporal reasoning, counting, and struggle to understand non-phonemic information, relying more on prompts than audio content. BLAB serves as a challenging evaluation framework to develop audio LMs with robust long-form audio understanding capabilities.

Eval3D: Interpretable and Fine-grained Evaluation for 3D Generation

Despite the unprecedented progress in the field of 3D generation, current systems still often fail to produce high-quality 3D assets that are visually appealing and geometrically and semantically consistent across multiple viewpoints. To effectively assess the quality of the generated 3D data, there is a need for a reliable 3D evaluation tool. Unfortunately, existing 3D evaluation metrics often overlook the geometric quality of generated assets or merely rely on black-box multimodal large language models for coarse assessment. In this paper, we introduce Eval3D, a fine-grained, interpretable evaluation tool that can faithfully evaluate the quality of generated 3D assets based on various distinct yet complementary criteria. Our key observation is that many desired properties of 3D generation, such as semantic and geometric consistency, can be effectively captured by measuring the consistency among various foundation models and tools. We thus leverage a diverse set of models and tools as probes to evaluate the inconsistency of generated 3D assets across different aspects. Compared to prior work, Eval3D provides pixel-wise measurement, enables accurate 3D spatial feedback, and aligns more closely with human judgments. We comprehensively evaluate existing 3D generation models using Eval3D and highlight the limitations and challenges of current models.

On Linear Representations and Pretraining Data Frequency in Language Models

Pretraining data has a direct impact on the behaviors and quality of language models (LMs), but we only understand the most basic principles of this relationship. While most work focuses on pretraining data's effect on downstream task behavior, we investigate its relationship to LM representations. Previous work has discovered that, in language models, some concepts are encoded `linearly' in the representations, but what factors cause these representations to form? We study the connection between pretraining data frequency and models' linear representations of factual relations. We find evidence that the formation of linear representations is strongly connected to pretraining term frequencies; specifically for subject-relation-object fact triplets, both subject-object co-occurrence frequency and in-context learning accuracy for the relation are highly correlated with linear representations. This is the case across all phases of pretraining. In OLMo-7B and GPT-J, we discover that a linear representation consistently (but not exclusively) forms when the subjects and objects within a relation co-occur at least 1k and 2k times, respectively, regardless of when these occurrences happen during pretraining. Finally, we train a regression model on measurements of linear representation quality in fully-trained LMs that can predict how often a term was seen in pretraining. Our model achieves low error even on inputs from a different model with a different pretraining dataset, providing a new method for estimating properties of the otherwise-unknown training data of closed-data models. We conclude that the strength of linear representations in LMs contains signal about the models' pretraining corpora that may provide new avenues for controlling and improving model behavior: particularly, manipulating the models' training data to meet specific frequency thresholds.

DataDecide: How to Predict Best Pretraining Data with Small Experiments

Because large language models are expensive to pretrain on different datasets, using smaller-scale experiments to decide on data is crucial for reducing costs. Which benchmarks and methods of making decisions from observed performance at small scale most accurately predict the datasets that yield the best large models? To empower open exploration of this question, we release models, data, and evaluations in DataDecide -- the most extensive open suite of models over differences in data and scale. We conduct controlled pretraining experiments across 25 corpora with differing sources, deduplication, and filtering up to 100B tokens, model sizes up to 1B parameters, and 3 random seeds. We find that the ranking of models at a single, small size (e.g., 150M parameters) is a strong baseline for predicting best models at our larger target scale (1B) (~80% of com parisons correct). No scaling law methods among 8 baselines exceed the compute-decision frontier of single-scale predictions, but DataDecide can measure improvement in future scaling laws. We also identify that using continuous likelihood metrics as proxies in small experiments makes benchmarks including MMLU, ARC, HellaSwag, MBPP, and HumanEval >80% predictable at the target 1B scale with just 0.01% of the compute.