Ola

Why is GPT-3 considered undertrained after Chinchilla?

Why does a small FT dramatically improve strict JSON output?

Walk through how a token's d_model embedding is split into per head pieces inside multi-head attention.

Does MoE reduce GPU memory footprint at inference compared to a dense model?

What is d_model, and which three parts of the architecture scale with it?

What should you do when naive safety FT makes the model refuse 'how does anesthesia work?'

Match each embedding model selection scenario to the model attribute or check that most determines the right choice.

What does dividing pre-softmax attention scores by an extra factor > 1 do at inference?

LLM-as-judge scores drift up across 50k weekly evals while user complaints hold steady. Diagnose likely causes and design a calibration system preventing silent score inflation.

Define recall@10 for a vector index, and say exactly what ground truth means in that definition.