Sergey Troshin

Sergey Troshin, Vlad Niculae, Antske Fokkens

Language models trained on large amounts of data are known to produce inappropriate content in some cases and require careful tuning to be used in the real world. We revisit an effective and modular approach for controllability of the language models, when an external expert model guides the decoding. Particularly, we zoom in into the parametrization choice of an external expert, highlighting the difference between low-rank and higher-rank parametrizations. Higher-rank experts are designed to support high flexibility when representing the rewards, leading to higher computational costs during decoding. However, we demonstrate that they might not use their full flexibility. By analyzing the recently proposed reward-augmented decoding approach (RAD), which uses a higher-rank expert model, we introduce a simpler but more efficient low-rank parametrization of the expert model enabling fast and effective guided decoding. We empirically show that the low-rank RAD performs on par with the more flexible RAD on a detoxification and a sentiment control task, while requiring only a single reward model call per generated token.

Evgeniia Tokarchuk, Maya K. Nachesa, Sergey Troshin et al.

Modern neural translation models based on the Transformer architecture are known for their high performance, particularly when trained on high-resource datasets. A standard next-token prediction training strategy, while widely adopted in practice, may lead to overlooked artifacts such as representation collapse. Previous works have shown that this problem is especially pronounced in the representation of the deeper Transformer layers, where it often fails to efficiently utilize the geometric space. Representation collapse is even more evident in end-to-end training of continuous-output neural machine translation, where the trivial solution would be to set all vectors to the same value. In this work, we analyze the dynamics of representation collapse at different levels of discrete and continuous NMT transformers throughout training. We incorporate an existing regularization method based on angular dispersion and demonstrate empirically that it not only mitigates collapse but also improves translation quality. Furthermore, we show that quantized models exhibit similar collapse behavior and that the benefits of regularization are preserved even after quantization.

Sergey Troshin, Wafaa Mohammed, Yan Meng et al.

Diversity is an essential metric for evaluating the creativity of outputs generated by language models. Temperature-based sampling is a common strategy to increase diversity. However, for tasks that require high precision, e.g., mathematical reasoning, uncontrolled high temperature sampling, e.g., min-$p$ or top-$p$, degrades reasoning quality. We demonstrate that the loss of accuracy is caused by sampling incorrect continuations in sensitive decoding positions. To address this, in this paper, we propose \textbf{selective sampling}, a method that dynamically switches between greedy and high-temperature sampling based on a sampling risk metric. This risk metric estimates the likelihood of output errors when applying high-temperature sampling on the current token position. To predict sampling risk, we train a lightweight classifier on a small subset of verifiable problems. The trained classifier can be integrated with the base language model with minimal latency overhead. Experiments on mathematical reasoning tasks demonstrate that selective sampling enhances the quality-diversity trade-off, even in high-temperature settings.

Sergey Troshin, Irina Saparina, Antske Fokkens et al.

Large language models increasingly rely on explicit reasoning chains and can produce multiple plausible responses for a given context. We study the candidate sampler that produces the set of plausible responses contrasting the ancestral (parallel) sampling against two alternatives: enumeration, which asks the model to produce $n$ candidates in one pass, and iterative sampling, which proposes candidates sequentially while conditioning on the currently generated response set. Under matched budgets, we compare these samplers on quality, lexical and computation flow diversity, and efficiency. Our empirical results demonstrate that enumeration and iterative strategies result in higher diversity at comparable quality. Our findings highlight the potential of simple non-independent sampling strategies to improve response diversity without sacrificing generation quality.

Evgeniia Tokarchuk, Sergey Troshin, Vlad Niculae

Augmenting neural machine translation with external memory at decoding time, in the form of k-nearest neighbors machine translation ($k$-NN MT), is a well-established strategy for increasing translation performance. $k$-NN MT retrieves a set of tokens that occurred in the most similar contexts recorded in a prepared data store, using hidden state representations of translation contexts as vector lookup keys. One of the main disadvantages of this method is the high computational cost and memory requirements. Since an exhaustive search is not feasible in large data stores, practitioners commonly use approximate $k$-NN MT lookup, yet even such algorithms are a bottleneck. In contrast to research directions seeking to accelerate $k$-NN MT by reducing data store size or the number of lookup calls, we pursue an orthogonal direction based on the performance properties of approximate $k$-NN MT lookup data structures. In particular, we propose to encourage angular dispersion of the neural hidden representations of contexts. We show that improving dispersion leads to better balance in the retrieval data structures, accelerating retrieval and slightly improving translations.