Chizoba Obasi blog

This is Part 2 of a two-part FlashAttention series.

0.

Mixture of Experts (MoE) is one of the main techniques used to scale modern language models without making every token pay the full computational cost of the model.

Attention mechanisms have evolved considerably to make transformer inference faster and more memory-efficient.

Almost all the algorithms/methods covered so far have been action-value methods (except gradient-bandit algorithms, Section 2.8).

Meta’s unified model for promptable image and video segmentation.

Transformers are a sequence-to-sequence model: given an input sequence, produce an output sequence.

Muon stands for MomentUm Orthogonalized by Newton-Schulz and was invented by Keller Jordan.

Earlier this year, I decided to force myself to read more.

Eligibility traces are one of the basic mechanisms of RL that unify and generalize TD and Monte Carlo (MC) methods.

Let’s dive into the control problem now with parametric approximation of the action-value function $\hat{q}(s, a, \mathbf{w}) \approx q_{*}(s, a)$, where $\mathbf{w} \in \mathbb{R}^d$ is a finite-dimensional weight vector.

Let’s discuss the extension of off-policy methods from the tabular case (Ch. 6 & 7) to function approximation.