\documentclass{article}

\usepackage[margin=1in]{geometry}

\usepackage{xcolor}
\usepackage{titlesec}
\usepackage[scaled=0.96]{helvet} % Sets up Helvetica font
\definecolor{mauve}{rgb}{0.58,0,0.82}

\titleformat{\section}[hang]{\color{mauve}\normalfont\sffamily\Large}{\thesection}{1em}{}
\titleformat{\subsection}[hang]{\color{mauve}\normalfont\sffamily\large}{\thesubsection}{1em}{}

\title{Homework 5}
\author{Your Name}
\date{Due date: November 14 at 2 PM}

\begin{document}
\maketitle

\section*{Question 1.1 (5 points)}

\subsection*{A}
Present the following values:
\begin{itemize}
    \item Validation perplexity on 512-len set:
    \item Validation perplexity on 2048-len set:
    \item Training time:
    \item GPU VRAM usage:
\end{itemize}

\subsection*{B}
Justify the difference, if any, between the perplexity on the two sets.: [Your answer here]

\section*{Question 1.2 (5 points)}

\subsection*{B}
Present the following values:
\begin{itemize}
    \item Validation perplexity on 512-len set:
    \item Validation perplexity on 2048-len set:
    \item Training time:
    \item GPU VRAM usage:
\end{itemize}

\subsection*{C}
Compare the results to the ones you achieved on the previous question. Support your comments by
explaining how speedups and memory reductions are achieved by Flash Attention: [Your answer here]

\section*{Question 1.3 (5 points)}

\subsection*{A}
\begin{itemize}
    \item Maximum batch size:
    \item Minimum batch size:
\end{itemize}


\subsection*{B}
Gradient Checkpointing:

\subsection*{C}
Present the following values:
\begin{itemize}
    \item Validation perplexity on 512-len set:
    \item Validation perplexity on 2048-len set:
    \item Training time:
    \item GPU VRAM usage:
\end{itemize}

\subsection*{D}
Perplexity behavior:

\subsection*{E}
Present the following values:
\begin{itemize}
    \item Validation perplexity on 512-len set:
    \item Validation perplexity on 2048-len set:
\end{itemize}

\noindent Compare and comment: 

\section*{Question 1.4 (5 points)}

\subsection*{A}
Describe at least three setups, including validation perplexities achieved by each one: [Your answer here]


\section*{Question 2.1 (2 points)}

\subsection*{A} [Your answer here] 

\subsection*{B} [Your answer here] 

\section*{Question 2.2 (3 points)}

\subsection*{A} [Your answer here] 

\subsection*{B} [Your answer here] 

\subsection*{C} [Your answer here] 

\section*{Question 3.1 (5 points)}

\subsection*{B} [Your answer here] [Attach your plot here]

\section*{Question 3.2 (2 points)}

[Your answer here] 

\section*{Question 3.3 (5 points)}

\subsection*{A} [Your answer here] 

\subsection*{B} [Your answer here] 

\subsection*{C} [Your answer here] 

\subsection*{D} [Your answer here] 


\section*{Question 4}

\subsection*{4.1}
[Describe your use of coding assistants, if any]

\subsection*{4.2}
[Include table of chatbot interactions, if any]

\subsection*{4.3}
[Describe any other use of Generative AI, if applicable]

\end{document}