Skip to content

Commit

Permalink
~
Browse files Browse the repository at this point in the history
  • Loading branch information
Kaiyu Yang committed Jun 26, 2023
1 parent c7edcb3 commit ea7b0a2
Showing 1 changed file with 1 addition and 1 deletion.
2 changes: 1 addition & 1 deletion index.html
Original file line number Diff line number Diff line change
Expand Up @@ -402,7 +402,7 @@ <h2 class="title is-3"><span class="dvima">ReProver: Retrieval-Augmented Theorem
<div class="rows is-centered ">
<div class="row is-full-width">
<h2 class="title is-3"><span class="dvima">Experiments</span></h2>
<p style="font-size: 125%">We use LeanDojo Benchmark to train and evaluate ReProver. During testing, the tactic generator is combined with best-first search to search for complete proofs. The table below shows the percentage of theorem proved within 10 minutes. The \(\texttt{novel_premises}\) spilt is much more challenging. On both data splits, ReProver outperforms Lean's built-in proof automation tactic (tidy), a baseline that generates tactics directly without retrieval, and another baseline using GPT-4 to generate tactics in a zero-shot manner.</p>
<p style="font-size: 125%">We use LeanDojo Benchmark to train and evaluate ReProver. During testing, the tactic generator is combined with best-first search to search for complete proofs. The table below shows the percentage of theorem proved within 10 minutes. The \(\texttt{novel_premises}\) spilt is much more challenging than the \(\texttt{random}\) split. On both data splits, ReProver outperforms Lean's built-in proof automation tactic (tidy), a baseline that generates tactics directly without retrieval, and another baseline using GPT-4 to generate tactics in a zero-shot manner.</p>
<br />
<img src="images/results.png" width="50%" class="interpolation-image" alt="" style="display: block; margin-left: auto; margin-right: auto"/>
</div>
Expand Down

0 comments on commit ea7b0a2

Please sign in to comment.