Krtin Kumar

Ashkan Alinejad, Krtin Kumar, Ali Vahdat

Question answering systems (QA) utilizing Large Language Models (LLMs) heavily depend on the retrieval component to provide them with domain-specific information and reduce the risk of generating inaccurate responses or hallucinations. Although the evaluation of retrievers dates back to the early research in Information Retrieval, assessing their performance within LLM-based chatbots remains a challenge. This study proposes a straightforward baseline for evaluating retrievers in Retrieval-Augmented Generation (RAG)-based chatbots. Our findings demonstrate that this evaluation framework provides a better image of how the retriever performs and is more aligned with the overall performance of the QA system. Although conventional metrics such as precision, recall, and F1 score may not fully capture LLMs' capabilities - as they can yield accurate responses despite imperfect retrievers - our method considers LLMs' strengths to ignore irrelevant contexts, as well as potential errors and hallucinations in their responses.

Krtin Kumar, Peyman Passban, Mehdi Rezagholizadeh et al.

Embedding matrices are key components in neural natural language processing (NLP) models that are responsible to provide numerical representations of input tokens.\footnote{In this paper words and subwords are referred to as \textit{tokens} and the term \textit{embedding} only refers to embeddings of inputs.} In this paper, we analyze the impact and utility of such matrices in the context of neural machine translation (NMT). We show that detracting syntactic and semantic information from word embeddings and running NMT systems with random embeddings is not as damaging as it initially sounds. We also show how incorporating only a limited amount of task-specific knowledge from fully-trained embeddings can boost the performance NMT systems. Our findings demonstrate that in exchange for negligible deterioration in performance, any NMT model can be run with partially random embeddings. Working with such structures means a minimal memory requirement as there is no longer need to store large embedding tables, which is a significant gain in industrial and on-device settings. We evaluated our embeddings in translating {English} into {German} and {French} and achieved a $5.3$x compression rate. Despite having a considerably smaller architecture, our models in some cases are even able to outperform state-of-the-art baselines.

Vasileios Lioutas, Ahmad Rashid, Krtin Kumar et al.

Word-embeddings are vital components of Natural Language Processing (NLP) models and have been extensively explored. However, they consume a lot of memory which poses a challenge for edge deployment. Embedding matrices, typically, contain most of the parameters for language models and about a third for machine translation systems. In this paper, we propose Distilled Embedding, an (input/output) embedding compression method based on low-rank matrix decomposition and knowledge distillation. First, we initialize the weights of our decomposed matrices by learning to reconstruct the full pre-trained word-embedding and then fine-tune end-to-end, employing knowledge distillation on the factorized embedding. We conduct extensive experiments with various compression rates on machine translation and language modeling, using different data-sets with a shared word-embedding matrix for both embedding and vocabulary projection matrices. We show that the proposed technique is simple to replicate, with one fixed parameter controlling compression size, has higher BLEU score on translation and lower perplexity on language modeling compared to complex, difficult to tune state-of-the-art methods.