You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
<p>We evaluate Librarian against state-of-the-art code agents on MiniCode across all three domains. The results demonstrate Librarian's effectiveness at creating compressed, correct refactorings.</p>
273
+
274
+
<figureclass="table-figure">
275
+
<tableclass="table-styled">
276
+
<thead>
277
+
<tr>
278
+
<th><strong>Agent</strong></th>
279
+
<th><strong>Tokens</strong></th>
280
+
<th><strong>CC</strong></th>
281
+
<th><strong>Pass %</strong></th>
282
+
<th><strong>MDL</strong></th>
283
+
<th><strong>MDL %</strong></th>
284
+
</tr>
285
+
</thead>
286
+
<tbody>
287
+
<tr>
288
+
<td>original</td>
289
+
<td>12946</td>
290
+
<td>238</td>
291
+
<td>82.0</td>
292
+
<td>13346.09</td>
293
+
<td>100.0</td>
294
+
</tr>
295
+
<tr>
296
+
<td>sonnet 3.7</td>
297
+
<td>18157</td>
298
+
<td>280</td>
299
+
<td>93.9</td>
300
+
<td>14357.77</td>
301
+
<td>107.4</td>
302
+
</tr>
303
+
<tr>
304
+
<td>sonnet 4</td>
305
+
<td>12756</td>
306
+
<td>224</td>
307
+
<td>84.4</td>
308
+
<td>10292.62</td>
309
+
<td>77.1</td>
310
+
</tr>
311
+
<tr>
312
+
<td>codex-mini</td>
313
+
<td>13178</td>
314
+
<td>238</td>
315
+
<td>82.0</td>
316
+
<td>11608.58</td>
317
+
<td>86.8</td>
318
+
</tr>
319
+
</tbody>
320
+
</table>
321
+
<figcaptionstyle="text-align: center;">Table 2: Results on the MiniCode CodeContests split</figcaption>
0 commit comments