Large language models perform well on tasks that have undergone fine-tuning of instructions, but their performance on completely unseen tasks is often less than ideal. To overcome the challenge of cross-task generalization, task-level LoRAs combination is proposed, which does not require training a model for new tasks. Instead, it learns the LoRA modules combination weights based on a small number of samples to form the task model. However, task-level LoRAs combination only utilizes a few task modules due to its reliance on the weight enumeration method, and it also ignores the specificity between different instances. Therefore, we proposed an instance-level LoRAs composition for cross-task generalization, which selects appropriate multiple task LoRA modules for each input instance and dynamically determines the composition weights. Our experiments on publicly available datasets show that our method outperforms the typical method, LoraHub, in 16 out of 27 tasks. We release the source code at https://github.com/noname822/iLoraComp.git

This paper presents a model architecture and training pipeline for attribute value extraction from search queries. The model uses weak labels generated from customer interactions to train a transformer-based NER model. A two-stage normalization process is then applied to deal with the problem of a large label space: first, the model output is normalized onto common generic attribute values, then it is mapped onto a larger range of actual product attribute values. This approach lets us successfully apply a transformer-based NER model to the extraction of a broad range of attribute values in a real-time production environment for e-commerce applications, contrary to previous research. In an online test, we demonstrate business value by integrating the model into a system for semantic product retrieval and ranking.

This paper describes the Johns Hopkins University (JHU) and Kyoto University submissions to the Speech Translation evaluation campaign at IWSLT2018. Our end-to-end speech translation systems are based on ESPnet and implements an attention-based encoder-decoder model. As comparison, we also experiment with a pipeline system that uses independent neural network systems for both the speech transcription and text translation components. We find that a transfer learning approach that bootstraps the end-to-end speech translation system with speech transcription system’s parameters is important for training on small datasets.