2024-03-08 15:33:44 Publish script update

app.py

@@ -234,8 +234,6 @@ Note that pause and resume time cost was not included in the "Cost Per Token" co
             """
         )
         general_plots[general_plots.plot_name == "execution_costs"].apply(display_filtered_plot, axis=1)
-    with gr.Tab("Summary metrics"):
-        summary_metrics_plots.apply(display_filtered_plot, axis=1)
     with gr.Tab("Context length and parameters count"):
         general_plots[general_plots.plot_name != "execution_costs"].apply(display_filtered_plot, axis=1)
         gr.Markdown(
@@ -247,7 +245,9 @@ A lot of models had to be omitted due to their developers not disclosing their p
 Mainly OpenAI's GPT models and Google's Palm 2.
 """
         )
-    with gr.Tab("Combined plots"):
+    with gr.Tab("Summary quality metrics"):
+        summary_metrics_plots.apply(display_filtered_plot, axis=1)
+    with gr.Tab("Comprehensive models comparison"):
         with gr.Row():
             choices = combined_plots.header
             choices = choices[choices.str.contains("for model")]
@@ -275,6 +275,8 @@ Mainly OpenAI's GPT models and Google's Palm 2.
 Radial plots are used to compare the most important aspects of each model researched on this board using single images.
 
 All values are normalized and scaled into 0.25 to 1 range, 0 is left for unknown values.
+
+Some metrics were reversed in order to make the plots more readable, for example "Fast execution" is `1 - execution_time` scaled to 0-1 range and moved 0.25 units up as mentioned above.
                     
 To compare the parameters more thoroughly use the filtering box on top of this page and inspect individual tabs.
 """)