update

app.py

@@ -58,6 +58,29 @@ def baseline_load_data(model,version,metrics):
     
     return df
 
+def update_dropdowns(setting, dataset, model):
+    updates = {
+        "setting": gr.update(interactive=True),
+        "dataset": gr.update(interactive=True),
+        "model": gr.update(interactive=True),
+    }
+    
+    if setting == "memorization":
+        updates["dataset"] = gr.update(value="news", interactive=False)
+        updates["model"] = gr.update(value="llama2-7B-chat_newsqa", interactive=False)
+    elif dataset == "books":
+        updates["setting"] = gr.update(value="RAG", interactive=False)
+        if model == "llama2-7B-chat_newsqa":
+            updates["model"] = gr.update(value="llama2-7B-chat-hf", interactive=True)
+    elif model == "llama2-7B-chat_newsqa":
+        updates["setting"] = gr.update(value="memorization", interactive=False)
+        updates["dataset"] = gr.update(value="news", interactive=False)
+    elif model != "llama2-7B-chat_newsqa":
+        updates["setting"] = gr.update(value="RAG", interactive=False)
+    
+    return updates["setting"], updates["dataset"], updates["model"]
+    
+
 def load_data(model, version, metrics):
     baseline_df = baseline_load_data(model, version, metrics)
     # now for every file in "versions/{model}-{version}/*.csv"
@@ -109,57 +132,62 @@ with demo:
     with gr.Tabs():
         with gr.TabItem("Leaderboard"):
             with gr.Row():
-                version_dropdown = gr.Dropdown(
-                    choices=["1%", "5%", "10%"],
-                    label="🔄 Select Forget Percentage",
-                    value="10%",
+                setting_dropdown = gr.Dropdown(
+                    choices = ["RAG", "memorization"],
+                    label="🔄 Select Setting",
+                    value="RAG",
                 )
-                model_dropdown = gr.Dropdown(
-                    choices=["llama", "phi"],
-                    label="🔄 Select Base Model",
-                    value="llama",
+                dataset_dropdown = gr.Dropdown(
+                    choices = ['news', 'books'],
+                    label="🔄 Select Dataset",
+                    value="news",
                 )
-            with gr.Row():
-                metrics_checkbox = gr.CheckboxGroup(
-                    label="Select Metrics",
-                    choices=["ROUGE", "Truth Ratio", "Prob."],
-                    value = ["ROUGE", "Truth Ratio", "Prob."],
-                )
-
-            with gr.Row():
-                search_bar = gr.Textbox(
-                    placeholder="Search for methods...",
-                    show_label=False,
+                model_dropdown = gr.Dropdown(
+                    choices=["llama2-7B-chat-hf", "llama2-70B-chat-hf", "dbrx-instruct", "llama2-7B-chat_newsqa"],
+                    label="🔄 Select Model",
+                    value="llama2-7B-chat-hf",
                 )
 
             leaderboard_table = gr.components.Dataframe(
-                value=load_data("llama", "10%", ["ROUGE", "Truth Ratio", "Prob."]),
+                value=load_data("RAG", "news", "llama2-7B-chat-hf"),
                 interactive=True,
                 visible=True,
             )
-
-            version_dropdown.change(
+            
+            setting_dropdown.change(
                 change_version,
-                inputs=[model_dropdown,version_dropdown,metrics_checkbox],
+                inputs=[dataset_dropdown,model_dropdown],
                 outputs=leaderboard_table
             )
-
-            model_dropdown.change(
+            
+            dataset_dropdown.change(
                 change_version,
-                inputs=[model_dropdown,version_dropdown,metrics_checkbox],
+                inputs=[setting_dropdown, model_dropdown],
                 outputs=leaderboard_table
             )
-
-            search_bar.change(
-                search_leaderboard,
-                inputs=[leaderboard_table, search_bar,metrics_checkbox],
+            
+            model_dropdown.change(
+                change_version,
+                inputs=[setting_dropdown, dataset_dropdown],
                 outputs=leaderboard_table
             )
+            
+            setting_dropdown.change(
+            update_dropdowns,
+            inputs=[setting_dropdown, dataset_dropdown, model_dropdown],
+            outputs=[setting_dropdown, dataset_dropdown, model_dropdown]
+            )
+            
+            dataset_dropdown.change(
+                update_dropdowns,
+                inputs=[setting_dropdown, dataset_dropdown, model_dropdown],
+                outputs=[setting_dropdown, dataset_dropdown, model_dropdown]
+            )
 
-            metrics_checkbox.change(
-                change_version,
-                inputs=[model_dropdown,version_dropdown,metrics_checkbox],
-                outputs=leaderboard_table
+            model_dropdown.change(
+                update_dropdowns,
+                inputs=[setting_dropdown, dataset_dropdown, model_dropdown],
+                outputs=[setting_dropdown, dataset_dropdown, model_dropdown]
             )
     
     with gr.Accordion("Submit a new model for evaluation"):

uploads.py

@@ -5,7 +5,7 @@ import datetime
 import pandas as pd
 
 
-RESULTS_PATH = "locuslab/CoTaEval_leaderboard"
+RESULTS_PATH = "boyiwei/CoTaEval_leaderboard"
 api = HfApi()
 TOKEN = os.environ.get("TOKEN", None)
 YEAR_VERSION = "2024"