Improve article 2204.10019 (image size - too small)

[dginev]

Exact location of issue

Images for Figures 1 and 2: https://ar5iv.labs.arxiv.org/html/2204.10019#S2.F1.g1

Problem details The images are way too small, likely a sizing interpretation problem.

[dginev]

Source was:

\begin{wrapfigure}{r}{0.5\textwidth}
        \begin{center}
            \vspace{0mm}
            \includegraphics[scale=0.32,clip=false,trim=107 120 100  52]{prompt_generator.png}
        \end{center}
        \vspace{8mm}
        \caption{The prompt generator architecture. A T5-base encoder~\citep{T5} receives trainable prompt tokens $p'$ and the input $x$, and a cross attention network implemented following~\cite{jaegle2021perceiver} translates its variable length output sequence into a fixed length input dependent prompt, $p(x)$. Blue indicates a "frozen", non-trained module; orange indicates a trained module.  \label{fig:prompt_generator}}\vspace{-0.5mm}
    \end{wrapfigure}