314,552 interview questions from 6,000+ companies.
Tests preference data collection design and annotation rubric quality for video alignment.
Tests SFT pipeline design and data mixture construction for instruction-following video generation.
Tests choosing and validating quantitative metrics for video generation quality at Genmo against human judgments.
Tests understanding of latent diffusion components and video-specific compression design.
Tests ability to derive diffusion objectives and distinguish DDPM vs DDIM training and inference behavior.
Tests systems design for high-throughput video ingestion and preprocessing at training scale.
Tests strategies for normalizing heterogeneous video inputs during pre-training.
Tests distributed training design choices and trade-offs for scaling large models.
Tests practical training techniques for fitting large video models into limited GPU memory.
Tests conceptual and practical understanding of flow matching relative to diffusion methods.
Tests mathematical CFG understanding and practical trade-offs for video generation quality and stability.
Tests ability to design robust alignment objectives that resist exploitation.
Tests architectural trade-off reasoning for text-to-video diffusion model backbones.
Tests ability to implement temporal attention layers for high-dimensional video tensors in PyTorch.
Tests preference optimization adaptation to diffusion and understanding of likelihood computation challenges.
Tests scaling-law application and compute allocation strategy for diffusion training.
Tests implementation choices for temporal consistency mechanisms in diffusion-based video models.
Tests end-to-end RLHF design for video generation including reward model formulation.
Tests techniques for temporal artifact reduction and style stability in video generation.
