Mistral¶
Screen / ML quiz¶
Transformer internals, attention, positional encodings, scaling laws, optimization (Adam, learning rate schedules), training instabilities, evaluation, and fine‑tuning methods for LLMs.
Coding + ML implementation round¶
Implementing core ML algorithms from scratch (e.g., K‑means, simple training loops) in Python or PyTorch, with attention to correctness and performance.
Deep‑dive research / systems rounds¶
Questions about training large models on multi‑GPU clusters (NCCL, OOM failures, parallelism strategies). Discussion of past research, papers you have written/read, and how you would improve Mistral‑style models.