You will be the first member of the ML team and will grow along with it. As such, your responsibilities will start in model improvement but grow with our internal team & open-source research. During your internship, you will:
- establish a coherent way to evaluate our models on code-related tasks that are relevant for actual usage
- fine-tune LLMs for code-related tasks (can't make promises about training from scratch)
- integrate or create rich new datasets to feed into your training experiments
- Work with other teammates to integrate the models you trained in Quack products
About you
As an engineer, you have a disturbing obsession with making something useful, not something shiny. Your curious nature makes you excited about modern technologies and tools (e.g. ChatGPT and/or GitHub Copilot are in your daily toolset). Your peers describe you as humble, and you make sure to always learn new things however experienced you may be. This drives you not to shy away from community/user feedback, but instead to seek the hard truth and iterate.
In short, you will probably be a good fit if you:
- have already founded or intend to found a startup at some point
- spend more time on your favorite newsfeed (e.g. Twitter, GitHub) browsing the latest models and research papers. Although given the choice, you'd prefer to share your model checkpoint and your code publicly, rather than spending weeks writing the research paper for citations.
- have experience with deep learning frameworks (PyTorch), NLP (Transformers, etc.) and GPUs (e.g. you've seem OOM & you know a few tricks to mitigate them).
- you have already trained or finetuned deep learning models (computer vision or NLP) and you're also comfortable running inference with LLMs (not 3rd party API).
- are comfortable using Python (more than Jupyter notebooks) and GitHub, you focus on results but you don't leave a mess behind. Open-source experience is plus
Please note we only consider internships onsite, for a period of 5-6 months! The internship is meant to lead to a full-time position.