Few-shot in-context learning (ICL) enables pre-trained language models to perform a previously-unseen task without any gradient-based training by feeding a small number of training examples as part of the input. ICL incurs substantial computational, memory, and storage costs because it involves processing all of the training examples every time a prediction is made. Parameter-efficient fine-tuning (e.g. adapter modules, prompt tuning, sparse update methods, etc.) offers an alternative paradigm where a small set of parameters are trained to enable a model to perform the new task. In this paper, we rigorously compare few-shot ICL and parameter-efficient fine-tuning and demonstrate that the latter offers better accuracy as well as dramatically lower computational costs. Along the way, we introduce a new parameter-efficient fine-tuning method called (IA) 3 that scales activations by learned vectors, attaining stronger performance while only introducing a relatively tiny amount of new parameters. We also propose a simple recipe based on the T0 model [1] called T-Few that can be applied to new tasks without task-specific tuning or modifications. We validate the effectiveness of T-Few on completely unseen tasks by applying it to the RAFT benchmark [2], attaining super-human performance for the first time and outperforming the state-of-the-art by 6% absolute. All of the code used in our experiments is publicly available. 1 * Equal contribution. 1 https://github.com/r-three/t-few Preprint. Under review.

Background: Coronavirus disease 2019 (COVID-19), a pandemic, infected millions of individuals globally. Several studies have been done on the demographic, clinical, laboratory, and radiological parameters of SARCoV-2 disease. However, information on the association of lipid profiles with disease severity is very sparse. SAR-CoV-2 patients with dysglycemia and hypertension are shown to have a worse prognosis. Dyslipidemia is considered one of the most important cardiovascular risk factors. The association of lipid profile with disease severity is less commonly studied. Although it is a cost-effective, easily performed, and routinely done test at clinical and hospital setups, it can be an independent risk factor for assessing disease severity. Aims and Objectives: The present study was undertaken to investigate lipid profile variation in SARS-CoV-2 infections and its association with disease severity. Materials and Methods: The Institutional Ethics Committee of Malla Reddy Institute of Medical Sciences and Hospital, Hyderabad, approved this retrospective study before it was conducted there. A total of 400 SARS-CoV-2 patients were enrolled in this study. From March 2, 2021 to December 4, 2021. All the data regarding serum lipid profiles were extracted electronically. Results: We found that mean values of total cholesterol (TC) (98.15), high-density lipoprotein cholesterol (HDL-C) (25.11), and low-density lipoprotein cholesterol (LDL-C) (38.30) compared to mild cases were significantly lower in severe cases: were significantly lower T-C (132.74), HDL-C (33.50), and LDL-C (59.25). A statistically significant difference was found between both groups in terms of TC (P < 0.001), HDL-C (P < 0.001), and LDL-C (P < 0.001). In addition, in the severe cases, triglycerides (TGs) levels showed an upward trend (256.95) and were significantly higher than in the mild cases (153.07) (P < 0.001). Conclusions: Decreased levels of TC, LDL-C, and HDL-C and increased levels of TGs were observed in severely ill Covid-19 cases. Higher TG levels immediately correlate with disease severity, and lower concentrations of TC, LDL-C, and HDL-C are reciprocally associated with it. Our study about lipid profiles helps provide a new perspective on SAR-CoV-2 infection.