Low-rank adaptation (LoRA) has become the default approach to fine-tune large language models (LLMs) due to its significant reduction in trainable parameters. However, trainable parameter demand for LoRA increases with increasing model embedding dimensions, leading to high compute costs. Additionally, its backward updates require storing high-dimensional intermediate activations and optimizer states, demanding high peak GPU memory. In this paper, we introduce _LaMDA_, a novel approach to fine-tuning large language models, which leverages low-dimensional adaptation to achieve significant reductions in trainable parameters and peak GPU memory footprint. LaMDA freezes a first projection matrix (PMA) in the adaptation path while introducing a low-dimensional trainable square matrix, resulting in substantial reductions in trainable parameters and peak GPU memory usage. LaMDA gradually freezes a second projection matrix (PMB) during the early fine-tuning stages, reducing the compute cost associated with weight updates to enhance parameter efficiency further.We also present an enhancement, LaMDA++, incorporating a “lite-weight” adaptive rank allocation for the LoRA path via normalized spectrum analysis of pre-trained model weights. We evaluate LaMDA/LaMDA++ across various tasks, including natural language understanding with the GLUE benchmark, text summarization, natural language generation, and complex reasoning on different LLMs.Results show that LaMDA matches or surpasses the performance of existing alternatives while requiring up to **17.7×** fewer parameter updates and up to **1.32×** lower peak GPU memory usage during fine-tuning. Code will be publicly available at https://github.com/ArminAzizi98/LaMDA.

We present a novel Parameter-Efficient Fine-Tuning (PEFT) method, dubbed as Adaptive Freezing of Low-Rank Adaptation (AFLoRA). Specifically, for each pre-trained frozen weight tensor, we add a parallel path of trainable low-rank matrices, namely a down-projection and an up-projection matrix, each of which is followed by a feature transformation vector. Based on a novel freezing score, we incrementally freeze these projection matrices during fine-tuning to reduce the computation and alleviate over-fitting. Our experimental results demonstrate that we can achieve state-of-the-art performance with an average improvement of up to 0.85% as evaluated on the GLUE benchmark while yielding up to 9.5× fewer average trainable parameters. While compared in terms of runtime, AFLoRA can yield up to 1.86× improvement as opposed to similar PEFT alternatives. Besides the practical utility of our approach, we provide insights on the trainability requirements of LoRA paths at different modules and the freezing schedule for the different projection matrices.

Due to the success of pre-trained language models, versions of languages other than English have been released in recent years. This fact implies the need for resources to evaluate these models. In the case of Spanish, there are few ways to systematically assess the models’ quality. In this paper, we narrow the gap by building two evaluation benchmarks. Inspired by previous work (Conneau and Kiela, 2018; Chen et al., 2019), we introduce Spanish SentEval and Spanish DiscoEval, aiming to assess the capabilities of stand-alone and discourse-aware sentence representations, respectively. Our benchmarks include considerable pre-existing and newly constructed datasets that address different tasks from various domains. In addition, we evaluate and analyze the most recent pre-trained Spanish language models to exhibit their capabilities and limitations. As an example, we discover that for the case of discourse evaluation tasks, mBERT, a language model trained on multiple languages, usually provides a richer latent representation than models trained only with documents in Spanish. We hope our contribution will motivate a fairer, more comparable, and less cumbersome way to evaluate future Spanish language models.

Ensuring smooth communication is essential in a chat-oriented dialogue system, so that a user can obtain meaningful responses through interactions with the system. Most prior work on dialogue research does not focus on preventing dialogue breakdown. One of the major challenges is that a dialogue system may generate an undesired utterance leading to a dialogue breakdown, which degrades the overall interaction quality. Hence, it is crucial for a machine to detect dialogue breakdowns in an ongoing conversation. In this paper, we propose a novel dialogue breakdown detection model that jointly incorporates a pretrained cross-lingual language model and a co-attention network. Our proposed model leverages effective word embeddings trained on one hundred different languages to generate contextualized representations. Co-attention aims to capture the interaction between the latest utterance and the conversation history, and thereby determines whether the latest utterance causes a dialogue breakdown. Experimental results show that our proposed model outperforms all previous approaches on all evaluation metrics in both the Japanese and English tracks in Dialogue Breakdown Detection Challenge 4 (DBDC4 at IWSDS2019).

Despite recent progress in conversational question answering, most prior work does not focus on follow-up questions. Practical conversational question answering systems often receive follow-up questions in an ongoing conversation, and it is crucial for a system to be able to determine whether a question is a follow-up question of the current conversation, for more effective answer finding subsequently. In this paper, we introduce a new follow-up question identification task. We propose a three-way attentive pooling network that determines the suitability of a follow-up question by capturing pair-wise interactions between the associated passage, the conversation history, and a candidate follow-up question. It enables the model to capture topic continuity and topic shift while scoring a particular candidate follow-up question. Experiments show that our proposed three-way attentive pooling network outperforms all baseline systems by significant margins.

We propose a novel, path-based reasoning approach for the multi-hop reading comprehension task where a system needs to combine facts from multiple passages to answer a question. Although inspired by multi-hop reasoning over knowledge graphs, our proposed approach operates directly over unstructured text. It generates potential paths through passages and scores them without any direct path supervision. The proposed model, named PathNet, attempts to extract implicit relations from text through entity pair representations, and compose them to encode each path. To capture additional context, PathNet also composes the passage representations along each path to compute a passage-based representation. Unlike previous approaches, our model is then able to explain its reasoning via these explicit paths through the passages. We show that our approach outperforms prior models on the multi-hop Wikihop dataset, and also can be generalized to apply to the OpenBookQA dataset, matching state-of-the-art performance.

Recently, there has been a surge of interest in reading comprehension-based (RC) question answering (QA). However, current approaches suffer from an impractical assumption that every question has a valid answer in the associated passage. A practical QA system must possess the ability to determine whether a valid answer exists in a given text passage. In this paper, we focus on developing QA systems that can extract an answer for a question if and only if the associated passage contains an answer. If the associated passage does not contain any valid answer, the QA system will correctly return Nil. We propose a novel nil-aware answer span extraction framework that is capable of returning Nil or a text span from the associated passage as an answer in a single step. We show that our proposed framework can be easily integrated with several recently proposed QA models developed for reading comprehension and can be trained in an end-to-end fashion. Our proposed nil-aware answer extraction neural network decomposes pieces of evidence into relevant and irrelevant parts and then combines them to infer the existence of any answer. Experiments on the NewsQA dataset show that the integration of our proposed framework significantly outperforms several strong baseline systems that use pipeline or threshold-based approaches.