playing around with gradio

app.py

@@ -6,35 +6,26 @@ import numpy as np
 import gradio as gr
 
 
-def sales_projections(employee_data):
-    sales_data = employee_data.iloc[:, 1:4].astype("int").to_numpy()
-    regression_values = np.apply_along_axis(
-        lambda row: np.array(np.poly1d(np.polyfit([0, 1, 2], row, 2))), 0, sales_data
-    )
-    projected_months = np.repeat(
-        np.expand_dims(np.arange(3, 12), 0), len(sales_data), axis=0
-    )
-    projected_values = np.array(
-        [
-            month * month * regression[0] + month * regression[1] + regression[2]
-            for month, regression in zip(projected_months, regression_values)
-        ]
-    )
-    plt.plot(projected_values.T)
-    plt.legend(employee_data["Name"])
-    return employee_data, plt.gcf(), regression_values
+dummy_data = [1, 2, 3, 4]
+
+
+def get_plot(model_name):
+    plt.plot(dummy_data)
+    plt.legend(model_name)
+    return plt.gcf()
 
 demo = gr.Blocks()
 
 with demo:
     with gr.Tabs():
         with gr.TabItem("Greedy Search"):
-            gr.Dropdown(
+            model_name = gr.Dropdown(
                 choices=["DistilGPT2", "GPT2", "OPT 1.3B", "GPTJ-6B", "T5 Small", "T5 Base", "T5 Large", "T5 3B"],
                 value="T5 Small",
                 label="Model",
                 interactive=True,
             )
+            get_plot(model_name)
         with gr.TabItem("Sample"):
             gr.Button("New Tiger")
         with gr.TabItem("Beam Search"):
@@ -48,6 +39,7 @@ with demo:
                     ["Pytorch Version", "1.11.0"],
                     ["OS", "22.04 LTS (3090) / Debian 10 (other GPUs)"],
                     ["CUDA", "11.6 (3090) / 11.3 (others GPUs)"],
+                    ["Number of runs", "100 (the first run was discarded to ignore compilation time)"],
                     ["Is there code to reproduce?", "Yes -- https://gist.github.com/gante/f0017e3f13ac11b0c02e4e4db351f52f"],
                 ],
             )