AI Researcher

Which of these reduce catastrophic forgetting in practice?

How do you detect catastrophic forgetting after a domain fine-tune?

Which signal is the most reliable indicator of catastrophic forgetting?

AdaLoRA: how does adaptive rank allocation across layers work?

Why does tree-based speculative decoding outperform sequential draft chains at the same target-pass cost?

How does vanilla speculative decoding work and why is verification a single forward pass?

Given draft acceptance rate α and target/draft cost ratio c, when does speculative decoding actually win?

Predict whether speculative decoding wins for given α, K, c

Apply the roofline model to LLM inference and identify where decode and prefill sit on it.

How does DeepSeek's Multi-Latent Attention (MLA) compress the KV cache below GQA?

How are RoPE NTK scaling and YaRN used at inference time to extend context beyond the training length?

For a 64-Q-head model with GQA group size G=8, how many K/V heads exist and what is the cache reduction vs MHA?

Why is 'reduce FLOPs to speed up decode' the most common beginner misconception in LLM serving?

Walk through the byte accounting that proves a single-batch decode step is bandwidth-bound on H100.

Predict the bandwidth-vs-compute latency of a single Llama-70B decode step on H100

Predict the critical batch size where decode crosses from memory-bound to compute-bound on H100

What is the arithmetic intensity of an LLM decode step and why is it close to 1?

Match each advanced speculative-decoding variant to its defining mechanism

How does WordPiece decide which pair to merge, compared to BPE?

How would you train a custom tokenizer for a genomics LLM on DNA sequences?

For a genomics tokenizer with a 4-character DNA alphabet, what vocabulary size range is likely sufficient?

How does vocabulary size affect embedding table memory footprint at bfloat16 precision, and what are the tradeoffs of very small vs. very large vocabularies?

Which part of a transformer model's parameter count scales directly and linearly with vocabulary size?

Explain the Unigram LM tokenization algorithm and how its training differs fundamentally from BPE. What does stochastic tokenization enable?

Which statement correctly describes the Unigram Language Model tokenizer's training procedure?