diff --git a/assets/swe_bench.jpg b/assets/swe_bench.jpg index 37c2769ce..5df496d66 100644 Binary files a/assets/swe_bench.jpg and b/assets/swe_bench.jpg differ diff --git a/assets/swe_bench.svg b/assets/swe_bench.svg index 3f5583403..d79ba8334 100644 --- a/assets/swe_bench.svg +++ b/assets/swe_bench.svg @@ -6,7 +6,7 @@ - 2024-05-31T09:30:46.920987 + 2024-05-31T11:28:28.622491 image/svg+xml @@ -30,8 +30,8 @@ z - - - + - + - + - + - + - + - + - + - + @@ -702,19 +702,19 @@ z - + - + - + @@ -722,7 +722,7 @@ z - + - + @@ -808,12 +808,12 @@ z - + - + - + - + @@ -999,7 +999,7 @@ z - + @@ -1012,12 +1012,12 @@ z - + - + - + - + - + @@ -1128,7 +1128,7 @@ z - + @@ -1137,7 +1137,7 @@ z - + @@ -1149,12 +1149,12 @@ z - + - + @@ -1162,7 +1162,7 @@ z - + @@ -1171,7 +1171,7 @@ z - + - + @@ -1240,23 +1240,23 @@ z - + - - + - + - + - + - + @@ -1294,18 +1294,18 @@ L 690 289.490154 - + - + - + @@ -1314,18 +1314,18 @@ L 690 254.819308 - + - + - + @@ -1334,18 +1334,18 @@ L 690 220.148462 - + - + - + - + - + - + @@ -1389,18 +1389,18 @@ L 690 150.80677 - + - + - + @@ -1410,18 +1410,18 @@ L 690 116.135924 - + - + - + @@ -1431,7 +1431,7 @@ L 690 81.465078 - + - - - @@ -1558,64 +1558,64 @@ L 690 50.4 " style="fill: none; stroke: #dddddd; stroke-width: 0.5; stroke-linejoin: miter; stroke-linecap: square"/> - +" clip-path="url(#pf552f9dc48)" style="fill: #b3d1e6; opacity: 0.3"/> - +" clip-path="url(#pf552f9dc48)" style="fill: #b3d1e6; opacity: 0.3"/> - +" clip-path="url(#pf552f9dc48)" style="fill: #b3d1e6; opacity: 0.3"/> - +" clip-path="url(#pf552f9dc48)" style="fill: #b3d1e6; opacity: 0.3"/> - +" clip-path="url(#pf552f9dc48)" style="fill: #b3d1e6; opacity: 0.3"/> - +" clip-path="url(#pf552f9dc48)" style="fill: #17965a; opacity: 0.9"/> - +" clip-path="url(#pf552f9dc48)" style="fill: #17965a; opacity: 0.9"/> - + @@ -1625,7 +1625,7 @@ z - + - + @@ -1677,7 +1677,7 @@ z - + - + @@ -1772,7 +1772,7 @@ z - + - + - + - + + diff --git a/benchmark/swe_bench_lite.py b/benchmark/swe_bench_lite.py index bbe949b67..0488c6bb7 100644 --- a/benchmark/swe_bench_lite.py +++ b/benchmark/swe_bench_lite.py @@ -86,7 +86,7 @@ def plot_swe_bench_lite(data_file): ) plt.figtext( 0.5, - 0.025, + 0.05, note, wrap=True, horizontalalignment="center", @@ -102,8 +102,6 @@ def plot_swe_bench_lite(data_file): imgcat(fig) ax.xaxis.label.set_color(font_color) - # Note: (570) and (2294) refer to the number of instances that were processed by the agent. - fname = sys.argv[1] plot_swe_bench_lite(fname)