Dynamically integrating new or rapidly evolving information after (Large) Language Model pre-training remains challenging, particularly in low-data scenarios or when dealing with private and specialized documents. In-context learning and retrieval-augmented generation (RAG) face limitations, including their high inference costs and their inability to capture global document information. In this paper, we propose a way of modularizing knowledge by training document-level Knowledge Modules (KMs). KMs are lightweight components implemented as parameter-efficient LoRA modules, which are trained to store information about new documents and can be easily plugged into models on demand. We show that next-token prediction performs poorly as the training objective for KMs. We instead propose Deep Context Distillation: we learn KMs parameters such as to simulate hidden states and logits of a teacher that takes the document in context. Our method outperforms standard next-token prediction and pre-instruction training techniques, across two datasets. Finally, we highlight synergies between KMs and retrieval-augmented generation.

We propose a grounded approach to meaning in language typology. We treat data from perceptual modalities, such as images, as a language-agnostic representation of meaning. Hence, we can quantify the function--form relationship between images and captions across languages. Inspired by information theory, we define "groundedness", an empirical measure of contextual semantic contentfulness (formulated as a difference in surprisal) which can be computed with multilingual multimodal language models. As a proof of concept, we apply this measure to the typology of word classes. Our measure captures the contentfulness asymmetry between functional (grammatical) and lexical (content) classes across languages, but contradicts the view that functional classes do not convey content. Moreover, we find universal trends in the hierarchy of groundedness (e.g., nouns > adjectives > verbs), and show that our measure partly correlates with psycholinguistic concreteness norms in English. We release a dataset of groundedness scores for 30 languages. Our results suggest that the grounded typology approach can provide quantitative evidence about semantic function in language.

The sparse Mixture-of-Experts (MoE) architecture is increasingly favored for scaling Large Language Models (LLMs) efficiently; however, MoE systems rely on heterogeneous compute and memory resources. These factors collectively influence the system's Cost, Accuracy, and Performance (CAP), creating a challenging trade-off. Current benchmarks often fail to provide precise estimates of these effects, complicating practical considerations for deploying MoE systems. To bridge this gap, we introduce MoE-CAP, a benchmark specifically designed to evaluate MoE systems. Our findings highlight the difficulty of achieving an optimal balance of cost, accuracy, and performance with existing hardware capabilities. MoE systems often necessitate compromises on one factor to optimize the other two, a dynamic we term the MoE-CAP trade-off. To identify the best trade-off, we propose novel performance evaluation metrics - Sparse Memory Bandwidth Utilization (S-MBU) and Sparse Model FLOPS Utilization (S-MFU) - and develop cost models that account for the heterogeneous compute and memory hardware integral to MoE systems. This benchmark is publicly available on HuggingFace: https://huggingface.co/spaces/sparse-generative-ai/open-moe-llm-leaderboard.

In-context learning (ICL) adapts LLMs by providing demonstrations without fine-tuning the model parameters; however, it does not differentiate between demonstrations and quadratically increases the complexity of Transformer LLMs, exhausting the memory. As a solution, we propose Mixtures of In-Context Learners (MoICL), a novel approach to treat subsets of demonstrations as experts and learn a weighting function to merge their output distributions based on a training set. In our experiments, we show performance improvements on 5 out of 7 classification datasets compared to a set of strong baselines (up to +13\% compared to ICL and LENS). Moreover, we enhance the Pareto frontier of ICL by reducing the inference time needed to achieve the same performance with fewer demonstrations. Finally, MoICL is more robust to out-of-domain (up to +11\%), imbalanced (up to +49\%), or noisy demonstrations (up to +38\%) or can filter these out from datasets. Overall, MoICL is a more expressive approach to learning from demonstrations without exhausting the context window or memory.

Parameter-efficient fine-tuning (PEFT) for cross-task generalization consists in pre-training adapters on a multi-task training set before few-shot adaptation to test tasks. Polytropon [Ponti et al., 2023] ($\texttt{Poly}$) jointly learns an inventory of adapters and a routing function that selects a (variable-size) subset of adapters for each task during both pre-training and few-shot adaptation. In this paper, we investigate the role that adapter routing plays in its success and design new variants based on our findings. First, we build on the intuition that finer-grained routing provides more expressivity. Hence, we propose $\texttt{MHR}$ (Multi-Head Routing) which combines subsets of adapter parameters and outperforms $\texttt{Poly}$ under a comparable parameter budget; by only fine-tuning the routing function and not the adapters ($\texttt{MHR}$-$z$) we achieve competitive performance with extreme parameter efficiency. Second, we find that $\texttt{Poly}$/$\texttt{MHR}$ performance is a result of better multi-task optimization, rather than modular inductive biases that facilitate adapter recombination and local adaptation, as previously hypothesized. In fact, we find that $\texttt{MHR}$ exhibits high gradient alignment between training tasks. We find that routing is most beneficial during multi-task pre-training rather than during few-shot adaptation and propose $\texttt{MHR}$-$\mu$, which discards routing and fine-tunes the average of the pre-trained adapters on each downstream tasks. This establishes $\texttt{MHR}$-$\mu$ as an effective method for single-adapter fine-tuning. We also show that $\texttt{MHR}$-$\mu$ can be used as an effective zero-shot transfer method by training the average of the pre-trained adapters for a few additional steps on the multi-task training set: this yields gains up to 3% on absolute accuracy w.r.t. the baselines.