Applied Research Internship - Retrieval Augmentation as Prompting

Here at Hugging Face, we’re on a journey to advance good Machine Learning and make it more accessible. Along the way, we contribute to the development of technology for the better.

We have built the fastest-growing, open-source library of pre-trained models in the world. With over 100M+ installs and 65K+ stars on GitHub, over 10 thousand companies are using HF technology in production, including leading AI organizations such as Google, Elastic, Salesforce, Algolia, and Grammarly.

About the Role

We propose a research project combining NLP and IR. The project's goal is to carry out research that will lead to results interesting to relevant research communities. During the course of the internship, we will work out a realistic way to communicate our conclusions to the broader community, e.g. via a research paper, a blog post, a demo, or an open-source repository. Below we describe a proposed topic for the research project.

Knowledge-Intensive NLP is often solved with retriever-reader architectures where first, a set of results relevant to a query is surfaced, then it is used to condition a language model which generates the final output (e.g. an answer to the input question).

If we look at this process in a retriever-agnostic way, we’re effectively doing prompting. The goal of this project is to study how we can prompt LLMs to elicit desired answers and use the findings to inform our understanding of the notion of relevance which then might be leveraged to build better retrievers. In particular, we’re interested in how redundancy in the results set (or in the prompt) impacts the language model’s predictions.

About You

• Given that this is a research intern project we are looking for candidates pursuing a PhD or MS level studies.
• The project will involve negotiating the research direction in collaboration with the intern mentor and other colleagues.
• Familiarity with research trends in NLP and IR, particularly retrieval augmentation in NLP, language modelling and prompting will be a big plus, the candidate will also be expected to read relevant literature.
• The project will require enough experience with Python to be able to build experiments and use existing research packages.
• Familiarity with the practice of training models on large-scale, GPU-provisioned research clusters will be a plus.

Preferred Location

Ideally, you are based in London, but we are open to remote work for the right candidate.

More about Hugging Face

We are actively working to build a culture that values diversity, equity, and inclusivity. We are intentionally building a workplace where people feel respected and supported—regardless of who you are or where you come from. We believe this is foundational to building a great company and community. Hugging Face is an equal opportunity employer and we do not discriminate on the basis of race, religion, color, national origin, gender, sexual orientation, age, marital status, veteran status, or disability status.

We value development. You will work with some of the smartest people in our industry. We are an organization that has a bias for impact and is always challenging ourselves to continuously grow. We provide all employees with reimbursement for relevant conferences, training, and education.

We care about your well-being. We offer flexible working hours and remote options. We support our employees wherever they are. While we have office spaces around the world, especially in the US, Canada, and Europe, we're very distributed and all remote employees have the opportunity to visit our offices. If needed, we'll also outfit your workstation to ensure you succeed.

We support the community. We believe significant scientific advancements are the result of collaboration across the field. Join a community supporting the ML/AI community.