Found the best Tesla interview questions

The best questions are the ones that have been asked before, so I'm sharing them here. I hope it helps to people that are applying for the AI infrastructure engineer position or other related job openings.<br><br>Here are a few questions asked by the interviewers:<br><br>a) Describe the architecture of a deep learning model (a simple one), how to train it, deploy it, and what are its tradeoffs.<br><br>b) What are the bottlenecks in data preparation, model training, and model inference.<br><br>c) What are the infrastructure requirements for parallel training.<br><br>d) Imagine you have a big HPC cluster, what are the bottlenecks that the scale of the cluster brings? How would you solve these bottlenecks?<br><br>e) How to improve the inference performance on the vehicle ( embodied to the vehicle, not on servers ).<br><br>f) What are the challenges of model training, specifically the matrix operations. How would you address these challenges.<br><br>g) What is your mental model of how AI tech scales. Walk me through it.<br><br>h) What are the bottlenecks in training a transformer-based model? How would you optimize these bottlenecks?<br><br>i) How do you evaluate the effectiveness of different scaling approaches for deep learning training? How do you decide when to scale horizontally versus vertically?<br><br>j) How would you architect a system for getting ground truth data?<br><br>k) Describe the process of how you would start to tackle a difficult technical infrastructure problem. What does your thought process look like for how to go about it?<br><br>l) Walk me through your reasoning process. What do you think about to come to a good answer or conclusion to a problem. Walk me through an example.<br><br>m) What is the significance (and any challenges) of engineering a model to fit in V100 Tensor Core? <br>How do you optimize its performance while training? Inference? <br>How do you come up with the block size for tensor core gemm? Yeah, block size isn’t really the right term. Yeah, there’s some memory matrix multiplication optimization that is happening…<br><br>n) How does gemm work? What are all the various ways to optimize it, and how is it actually optimized by the Titan V.<br><br>o) How do you deal with data coming from different datasets and how do you handle this data through the training process. E.g. how do you handle labels? And how do you handle data coming from the vehicle.<br><br>p) If you had to break down the challenges faced in obtaining training data by dataset, what are they? What approaches would you take to address these challenges? What are the advantages/disadvantages of each approach?<br><br>q) There are often trade-offs in deep learning methods, for example, time to train vs. performance. Other examples might be memory usage vs. performance, quality of training data vs. amount of training data. How do you evaluate what trade-offs should be made in a particular problem, and what are the important factors in making this evaluation?<br><br>r) The model training process is extremely compute and memory intensive. How do you address the amount of memory needed for multi head self attention?<br><br>s) How do you address the amount of memory needed for multi head self attention? What are the trade-offs between the various approaches?<br><br>t) Do you think Tesla is the best place for you to work? Why? Why do you want to work here?<br><br>u) Why are you a strong fit for the position? <br>Why are you the best candidate?<br><br>v) Have you ever worked in an environment where there is a lot of ambiguity, or changing plans. How do you function in an environment of ambiguity?<br><br>w) Why do you want to work at Tesla. Why do you want to do this job. What do you hope to get out of it?<br><br>x) Do you have any questions for me?<br><br>y) What are your long-term goals and objectives? Where do you see yourself 5 years from now. What do you hope to accomplish in your career?<br><br>z) Why do you want to work at Tesla? Why do you want to do this work? What do you hope to get out of it?<br><br>aa) Final comments?