[Add] load data from oLLMLv2, drawdown menus

app.py

@@ -1,28 +1,47 @@
-from datasets import load_dataset
-import numpy as np
-import matplotlib.pyplot as plt
 import gradio as gr
 
-def compute_similarity(dataset_name):
-    # Load dataset
-    #dataset = load_dataset(dataset_name)
-    # Dummy similarity computation (replace with your metric)
-    data = np.random.rand(10, 10)
-    # Create heatmap
-    fig, ax = plt.subplots()
-    cax = ax.matshow(data, cmap='viridis')
-    plt.colorbar(cax)
-    return fig
+from src.dataloading import get_leaderboard_models, get_leaderboard_datasets
 
-with gr.Blocks() as demo:
-    dataset_name = gr.Textbox(label="Enter Dataset Name (e.g., 'imdb')")
-    heatmap_plot = gr.Plot(label="Similarity Heatmap")
-    compute_button = gr.Button("Compute Similarity")
 
-    compute_button.click(
-        fn=compute_similarity,
-        inputs=dataset_name,
-        outputs=heatmap_plot
-    )
+def create_demo():
+    # Fetch data once on startup (cache this in production)
+    models = get_leaderboard_models()
+    datasets = get_leaderboard_datasets()
 
-demo.launch()
+    with gr.Blocks(title="LLM Similarity Analyzer") as demo:
+        gr.Markdown("## Compare Models/Datasets from Open LLM Leaderboard")
+        
+        with gr.Row():
+            model_dropdown = gr.Dropdown(
+                choices=models,
+                label="Select Model",
+                filterable=True,
+                interactive=True,
+                allow_custom_value=False,
+                info="Search models from Open LLM Leaderboard"
+            )
+            
+            dataset_dropdown = gr.Dropdown(
+                choices=datasets,
+                label="Select Dataset",
+                filterable=True,
+                interactive=True,
+                info="Leaderboard benchmark datasets"
+            )
+
+        # Add your similarity computation and visualization components here
+        # Example placeholder:
+        similarity_output = gr.Textbox(label="Similarity Score")
+        compute_btn = gr.Button("Compute Similarity")
+        
+        def compute_similarity(model, dataset):
+            # Replace with your actual similarity metric
+            return f"Similarity between {model} and {dataset}: {0.85:.2f}"
+        
+        compute_btn.click(
+            fn=compute_similarity,
+            inputs=[model_dropdown, dataset_dropdown],
+            outputs=similarity_output
+        )
+        
+    return demo

src/dataloading.py

@@ -0,0 +1,31 @@
+from huggingface_hub import HfApi
+
+def get_leaderboard_models():
+    api = HfApi()
+    
+    # List all files in the "open_llm_leaderboard" directory of the Space
+    files = api.list_files_info(
+        repo_id="open-llm-leaderboard/open_llm_leaderboard",
+        path="open_llm_leaderboard",
+        repo_type="space"
+    )
+    
+    models = []
+    for file in files:
+        if "-details" in file.path and "__" in file.path:
+            # Extract provider and model name from filename
+            filename = file.path.split("/")[-1].replace("-details", "")
+            provider, model = filename.split("__", 1)
+            models.append(f"{provider}/{model}")
+    
+    return sorted(list(set(models)))  # Remove duplicates
+
+def get_leaderboard_datasets():
+    return [
+        "ai2_arc",
+        "hellaswag",
+        "mmlu",
+        "truthful_qa",
+        "winogrande",
+        "gsm8k"
+    ]