Sebastian Jaszczur

Szymon Antoniak, Michał Krutul, Maciej Pióro et al.

Mixture of Experts (MoE) models based on Transformer architecture are pushing the boundaries of language and vision tasks. The allure of these models lies in their ability to substantially increase the parameter count without a corresponding increase in FLOPs. Most widely adopted MoE models are discontinuous with respect to their parameters - often referred to as sparse. At the same time, existing continuous MoE designs either lag behind their sparse counterparts or are incompatible with autoregressive decoding. Motivated by the observation that the adaptation of fully continuous methods has been an overarching trend in deep learning, we develop Mixture of Tokens (MoT), a simple, continuous architecture that is capable of scaling the number of parameters similarly to sparse MoE models. Unlike conventional methods, MoT assigns mixtures of tokens from different examples to each expert. This architecture is fully compatible with autoregressive training and generation. Our best models not only achieve a 3x increase in training speed over dense Transformer models in language pretraining but also match the performance of state-of-the-art MoE architectures. Additionally, a close connection between MoT and MoE is demonstrated through a novel technique we call transition tuning.

Jakub Krajewski, Jan Ludziejewski, Kamil Adamczewski et al.

Mixture of Experts (MoE) models have emerged as a primary solution for reducing the computational cost of Large Language Models. In this work, we analyze their scaling properties, incorporating an expanded range of variables. Specifically, we introduce a new hyperparameter, granularity, whose adjustment enables precise control over the size of the experts. Building on this, we establish scaling laws for fine-grained MoE, taking into account the number of training tokens, model size, and granularity. Leveraging these laws, we derive the optimal training configuration for a given computational budget. Our findings not only show that MoE models consistently outperform dense Transformers but also highlight that the efficiency gap between dense and MoE models widens as we scale up the model size and training budget. Furthermore, we demonstrate that the common practice of setting the size of experts in MoE to mirror the feed-forward layer is not optimal at almost any computational budget.

Jan Małaśnicki, Kamil Ciebiera, Mateusz Boruń et al.

Recent years have seen a growing interest and adoption of LLMs, with Mixture-of-Experts (MoE) emerging as a leading architecture in extremely large models. Currently, the largest open-source models reach over $1$T parameters. At such scales, hyperparameter tuning becomes prohibitively expensive. Precisely for this reason, the $μ$Transfer is becoming a key technique. It allows for seamless transfer of optimal hyperparameters across model scales, resulting in a huge reduction in tuning costs. However, existing work has primarily focused on dense LLMs, leaving MoE architectures unexplored. In this work, we derive a $μ$-Parameterization for MoE, providing theoretical guarantees for feature learning across model widths. Our experiments demonstrate that the optimal learning rate reliably transfers across model sizes, establishing a foundation for efficient hyperparameter tuning in large-scale MoE models.

Maciej Pióro, Kamil Ciebiera, Krystian Król et al.

State Space Models (SSMs) have become serious contenders in the field of sequential modeling, challenging the dominance of Transformers. At the same time, Mixture of Experts (MoE) has significantly improved Transformer-based Large Language Models, including recent state-of-the-art open models. We propose that to unlock the potential of SSMs for scaling, they should be combined with MoE. We showcase this on Mamba, a recent SSM-based model that achieves remarkable performance. Our model, MoE-Mamba, outperforms both Mamba and baseline Transformer-MoE. In particular, MoE-Mamba reaches the same performance as Mamba in $2.35\times$ fewer training steps while preserving the inference performance gains of Mamba against Transformer.

Konrad Staniszewski, Szymon Tworkowski, Sebastian Jaszczur et al.

Recent advancements in long-context large language models have attracted significant attention, yet their practical applications often suffer from suboptimal context utilization. This study investigates structuring training data to enhance semantic interdependence, demonstrating that this approach effectively improves context utilization. To this end, we introduce the Structured Packing for Long Context (SPLiCe) method, which utilizes retrieval to collate mutually relevant documents into long and coherent training examples. We validate SPLiCe empirically across models of varying sizes -- 3B, 7B, and 13B -- achieving improved performance in long-context tasks, such as Qasper and HotpotQA. Remarkably, even brief fine-tuning with SPLiCe is sufficient to realize these benefits. Additionally, SPLiCe effectively mitigates the lost-in-middle phenomenon often observed in large models. Our comprehensive analysis of SPLiCe explores its design choices and reveals intriguing transfer effects; for instance, training on programming code enhances performance on natural language tasks.

Jan Ludziejewski, Maciej Pióro, Jakub Krajewski et al.

Mixture of Experts (MoE) architectures have significantly increased computational efficiency in both research and real-world applications of large-scale machine learning models. However, their scalability and efficiency under memory constraints remain relatively underexplored. In this work, we present joint scaling laws for dense and MoE models, incorporating key factors such as the number of active parameters, dataset size, and the number of experts. Our findings provide a principled framework for selecting the optimal MoE configuration under fixed memory and compute budgets. Surprisingly, we show that MoE models can be more memory-efficient than dense models, contradicting conventional wisdom. To derive and validate the theoretical predictions of our scaling laws, we conduct over 280 experiments with up to 2.7B active parameters and up to 5B total parameters. These results offer actionable insights for designing and deploying MoE models in practical large-scale training scenarios.

Jan Ludziejewski, Jan Małaśnicki, Maciej Pióro et al.

In this work, we introduce a novel approach for optimizing LLM training by adjusting learning rates across weights of different components in Transformer models. Traditional methods often apply a uniform learning rate across all network layers, potentially overlooking the unique dynamics of each part. Remarkably, our introduced relative learning rates, RLRS, method accelerates the training process by up to $23\%$, particularly in complex models such as Mixture of Experts (MoE). Hyperparameters of RLRS can be efficiently tuned on smaller models and then effectively reused on models up to $27\times$ larger. This simple and effective method results in a substantial reduction in training time and computational resources, offering a practical and scalable solution for optimizing large-scale neural networks.

Maciej Stefaniak, Michał Krutul, Jan Małaśnicki et al.

Large language models have steadily increased in size to achieve improved performance; however, this growth has also led to greater inference time and computational demands. Consequently, there is rising interest in model size reduction methods. To address this issue, we propose Projected Compression, a novel model compression technique, that reduces model weights by utilizing projection modules. Specifically, we first train additional trainable projections weights and preserve access to all the original model parameters. Subsequently, these projections are merged into a lower-dimensional product matrix, resulting in a reduced-size standard Transformer-based model. Unlike alternative approaches that require additional computational overhead, our method matches the base model's per-token computation step in FLOPs. Experimental results show that Projected Compression outperforms the comparable hard pruning and retraining approach on higher quality models. Moreover, the performance margin scales well with the number of tokens.

Sebastian Jaszczur, Aakanksha Chowdhery, Afroz Mohiuddin et al.

Large Transformer models yield impressive results on many tasks, but are expensive to train, or even fine-tune, and so slow at decoding that their use and study becomes out of reach. We address this problem by leveraging sparsity. We study sparse variants for all layers in the Transformer and propose Scaling Transformers, a family of next generation Transformer models that use sparse layers to scale efficiently and perform unbatched decoding much faster than the standard Transformer as we scale up the model size. Surprisingly, the sparse layers are enough to obtain the same perplexity as the standard Transformer with the same number of parameters. We also integrate with prior sparsity approaches to attention and enable fast inference on long sequences even with limited memory. This results in performance competitive to the state-of-the-art on long text summarization.

Sebastian Jaszczur, Michał Łuszczyk, Henryk Michalewski

We use neural graph networks with a message-passing architecture and an attention mechanism to enhance the branching heuristic in two SAT-solving algorithms. We report improvements of learned neural heuristics compared with two standard human-designed heuristics.