Updated benchmark with latest data from MMMU

valentinfrlch · Oct 20, 2024 · cb495f6 · cb495f6
1 parent 07af2f8
commit cb495f6
Show file tree

Hide file tree

Showing 3 changed files with 2 additions and 2 deletions.
diff --git a/benchmark_visualization/benchmark_data.csv b/benchmark_visualization/benchmark_data.csv
@@ -1,7 +1,7 @@
 Model,Size,Date,Overall,Art & Design,Business,Science,Health & Medicine,Human. & Social Sci.,Tech & Eng.,Cost
 GPT-4o,-,2024-05-27,69.1,-,-,-,-,-,-,5
 GPT-4o mini,-,2024-05-27,59.4,-,-,-,-,-,-,0.15
-Gemini 1.5 Pro,-,2024-05-31,62.2,-,-,-,-,-,-,3.5
+Gemini 1.5 Pro,-,2024-05-31,65.8,-,-,-,-,-,-,3.5
 Gemini 1.0 Ultra,-,2023-12-11,59.4,70,56.7,48,67.3,78.3,47.1,
 Claude 3 Opus,-,2024-03-05,59.4,67.5,67.2,48.9,61.1,70,50.6,15
 Claude 3.5 Sonnet,-,2024-03-05,68.3,-,-,-,-,-,-,3

diff --git a/benchmark_visualization/benchmark_visualization.jpg b/benchmark_visualization/benchmark_visualization.jpg
diff --git a/blueprints/camera_motion_summary.yaml b/blueprints/camera_motion_summary.yaml
@@ -94,7 +94,7 @@ blueprint:
     detail:
       name: Detail
       description: Detail parameter (OpenAI only)
-      default: 'high'
+      default: 'low'
       selector:
         select:
           options: