The number of tokens it takes to encode parallel text in different languages is known to vary. These disparities are called token premiums. Having high token premiums leads to less throughput during training and increases costs at inference. In this paper, we show that even after controlling for dataset size, vocabulary size, and data content, monolingual tokenizers exhibit a wide range of token premiums across languages. To understand the cross-linguistic differences that cause these token premiums, we train a suite of approximately 7,000 comparable monolingual tokenizers for 97 languages, manipulating tokenization algorithm, vocabulary size, and dataset size.

The widespread use of diffusion models has led to an abundance of AI-generated data, raising concerns about model collapse--a phenomenon in which recursive iterations of training on synthetic data lead to performance degradation. Prior work primarily characterizes this collapse via variance shrinkage or distribution shift, but these perspectives miss practical manifestations of model collapse. This paper identifies a transition from generalization to memorization during model collapse in diffusion models, where models increasingly replicate training data instead of generating novel content during iterative training on synthetic samples. This transition is directly driven by the declining entropy of the synthetic training data produced in each training cycle, which serves as a clear indicator of model degradation. Motivated by this insight, we propose an entropy-based data selection strategy to mitigate the transition from generalization to memorization and alleviate model collapse. Empirical results show that our approach significantly enhances visual quality and diversity in recursive generation, effectively preventing collapse.

A.1 Hyperparameters for QLORA We do a hyperparameter search for LoRA over the following variables: LoRA dropout { 0.0, 0.05, 0.1}, LoRA r { 8, 16, 32, 64, 128, 256}, LoRA layers {key+query, all attention layers, all FFN layers, all layers, attention + FFN output layers}. We keep LoRA α fixed and search the learning rate, since LoRA α is always proportional to the learning rate. We find that LoRA dropout 0.05 is useful for small models (7B, 13B), but not for larger models (33B, 65B). Each dot represents a combination of hyperparameters and for each LoRA r we run 3 random seed with each hyperparameter combination. The performance of specific LoRA r values appears to be independent of other hyperparameters.

Scaling law that rewards large datasets, complex models and enhanced data granularity has been observed in various fields of deep learning.

To evaluate new VLMs that expand beyond the already implemented 59 VLMs, users need to follow Code Snippet 2. Users would need to create a class that inherent from As described in Section 2.2, LLM-style models defined as models that generate tokens/text as output. Thereby, making them hard to compare with CLIP-style VLMs. Following Matsuura et al. [2023] methodology, we evaluated Llava 1.5 [Liu et al., 2023] - a LLM-style VLM - on various benchmark types in UniBench (Table 2). Scaling improves many benchmarks, but offers little benefit for reasoning and relation. Figure 8: Benchmark capabilities performance does not scale with dataset and model size Median zero-shot performance of models on various benchmark capabilities.