SUSYUSTC / MathTranslate

translate scientific papers in latex, especially arxiv papers
https://github.com/SUSYUSTC/MathTranslate
Apache License 2.0
1.04k stars 69 forks source link

Error found in Parapragh 63 #64

Closed Ethan-Chen-plus closed 8 months ago

Ethan-Chen-plus commented 11 months ago

image

translate_arxiv.exe 2307.08621v4 -eng google
The current mathtranslate is latest
Start
engine google
language from en
language to zh-CN
threads auto

arxiv number: 2307.08621v4

temporary directory C:\Users\25122\AppData\Local\Temp\tmphwijv6j4
trying to download from https://arxiv.org/e-print/2307.08621v4
main tex files found:
.\main.tex
merging .\settings.tex
merging .\math_commands.tex
Processing .\main
It is a full latex document
  3%|██▊                                                                                | 3/90 [00:05<02:55,  2.01s/it]Error found in ParapraghError found in Parapragh  3232

Content
Content  \begin{table*}[t]
\begin{tabular}{lcccc}
\toprule
\bf Architectures      & \bf \tabincell{c}{Training  XMATHXBS  Parallelization} & \bf  Inference Cost & \bf \tabincell{c}{Long-Sequence  XMATHXBS  Memory Complexity} & \bf  Performance  XMATHXBS
\midrule
Transformer        & \ding{52} &   $O(N)$   &   $O(N^2)$   & \ding{52}\ding{52}  XMATHXBS
Linear Transformer & \ding{52} &   $O(1)$   &   $O(N)$   & \ding{56}  XMATHXBS
Recurrent NN       & \ding{56} &   $O(1)$   &   $O(N)$   & \ding{56}  XMATHXBS
RWKV               & \ding{56} &   $O(1)$   &   $O(N)$   & \ding{52}  XMATHXBS
H3/S4              & \ding{52} &   $O(1)$   &   $O(N \log N)$   & \ding{52}  XMATHXBS
Hyena              & \ding{52} &   $O(N)$   &   $O(N\log N)$   & \ding{52}  XMATHXBS
\our{}                & \ding{52} &   $O(1)$   &   $O(N)$   & \ding{52}\ding{52}  XMATHXBS
\bottomrule
\end{tabular}
\caption{Model comparison from various perspectives. \our{} achieves training parallelization, constant inference cost, linear long-sequence memory complexity, and good performance.}
\label{tbl:compare:aspects}
\end{table*}

  \paragraph{Inference}
The recurrent representation (  \Eqref{eq:ret:recurrent}  ) is employed during the inference, which nicely fits autoregressive decoding.
The   $O(1)$   complexity reduces memory and inference latency while achieving equivalent results.
  6%|████▌                                                                              | 5/90 [00:07<01:57,  1.38s/it]Error found in Parapragh 46
SUSYUSTC commented 11 months ago

我这边试了一下好像没有问题?

sherrylixuecheng commented 8 months ago

We thank your report! Since this issue is resolved, we close this issue :)