Is this a reasonable ML interview question
Is it a reasonable machine learning interview question to implement self-attention from scratch? Time limit: 20 minutes. Position: ML Engineer
I had this question at two different interviews. It’s super popular today and very reasonable. If you can’t answer it, then you probably are not the type of candidate they are looking for. Maybe you want to target MLops roles
Do you remember the formula? Coding is not a problem.
Do you mind sharing for what roles you had this question in two different interviews? Also: when you say that it is "very reasonable" could I trouble you to explain why you think so? I can't imagine this kind of question being reasonable or relevant to actual work that one would be doing on the job, unless it was an academic organization, or a heavily-research focused org where you are developing models completely from scratch.
Likely a popular question since you are never going to do it on the job
You will never, ever do this in an industry setting. I agree.
Hi can i get a referral op?
I would love to, and I know this is confusing bc of my handle on Blind...I used to work for Disney. I work for Amazon now, and on a whim I'm interviewing there (Disney) maybe to return if things work out and the offer is good.
Very easy Initialize 3 matrices k,q,v with random values Create the input matrix and multiply and add softmax the final shape should match the first 5 minutes max
suppose you can't use a library like numpy. Still easy in 20 minutes?
Easier if you use list of lists