update space

app.py

@@ -3,6 +3,9 @@ from gradio_leaderboard import Leaderboard, ColumnFilter, SelectColumns
 import pandas as pd
 import os
 import json
+import tempfile
+import shutil
+import zipfile
 from huggingface_hub import snapshot_download
 
 # Constants for PhysicalCodeBench
@@ -72,7 +75,7 @@ SUBMISSION_TEXT = """
    ├── evaluation_results/   # Directory containing all result files
    └── PhysCodeEval_results.json  # Main evaluation results file
    ```
-5. Submit a pull request with your results
+5. Submit your results by uploading a ZIP file below or via the form
 
 Your submission will be verified and added to the leaderboard once approved.
 """
@@ -111,7 +114,7 @@ COLUMNS = [
     PhysCodeColumn("rank", "number", True, True, False),
     PhysCodeColumn("model", "str", True, True, False),
     PhysCodeColumn("model_type", "str", True, False, False),
-    #PhysCodeColumn("params", "number", True, False, False),
+    PhysCodeColumn("organization", "str", True, False, False),
     PhysCodeColumn("text_score", "number", True, False, False),
     PhysCodeColumn("visual_score", "number", True, False, False),
     PhysCodeColumn("total_score", "number", True, False, False),
@@ -120,8 +123,7 @@ COLUMNS = [
     PhysCodeColumn("execution_success", "number", False, False, False),
     PhysCodeColumn("file_generation", "number", False, False, False),
     PhysCodeColumn("file_quality", "number", False, False, False),
-    PhysCodeColumn("submission_date", "date", False, False, False),
-    PhysCodeColumn("license", "str", False, False, False)
+    PhysCodeColumn("submission_date", "date", False, False, False)
 ]
 
 # Enums for model metadata
@@ -144,7 +146,7 @@ def get_leaderboard_df():
             "rank": 1,
             "model": "GPT4o",
             "model_type": ModelType.CloseSource,
-            "params": 1000,
+            "organization": "OpenAI",
             "text_score": 16.0,
             "visual_score": 18.262,
             "total_score": 34.262,
@@ -153,14 +155,13 @@ def get_leaderboard_df():
             "execution_success": 10.0,
             "file_generation": 3.0,
             "file_quality": 3.0,
-            "submission_date": "2025-01-15",
-            "license": "Proprietary"
+            "submission_date": "2025-01-15"
         },
         {
             "rank": 2,
             "model": "Gemini-2.0-flash",
             "model_type": ModelType.CloseSource,
-            "params": 450,
+            "organization": "Google",
             "text_score": 15.0,
             "visual_score": 16.963,
             "total_score": 31.963,
@@ -169,14 +170,13 @@ def get_leaderboard_df():
             "execution_success": 9.0,
             "file_generation": 3.0,
             "file_quality": 3.0,
-            "submission_date": "2025-01-20",
-            "license": "Proprietary"
+            "submission_date": "2025-01-20"
         },
         {
             "rank": 3,
             "model": "DS-R1",
             "model_type": ModelType.OpenSource,
-            "params": 32,
+            "organization": "DeepSeek",
             "text_score": 14.0,
             "visual_score": 15.815,
             "total_score": 29.815,
@@ -185,14 +185,13 @@ def get_leaderboard_df():
             "execution_success": 8.5,
             "file_generation": 3.0,
             "file_quality": 2.5,
-            "submission_date": "2025-01-25",
-            "license": "Apache 2.0"
+            "submission_date": "2025-01-25"
         },
         {
             "rank": 4,
             "model": "DeepSeek-R1-Distill-Qwen-32B",
             "model_type": ModelType.OpenSource,
-            "params": 32,
+            "organization": "DeepSeek",
             "text_score": 12.2,
             "visual_score": 15.82,
             "total_score": 28.02,
@@ -201,14 +200,13 @@ def get_leaderboard_df():
             "execution_success": 7.2,
             "file_generation": 2.5,
             "file_quality": 2.5,
-            "submission_date": "2025-01-28",
-            "license": "Apache 2.0"
+            "submission_date": "2025-01-28"
         },
         {
             "rank": 5,
             "model": "QwQ-32B",
             "model_type": ModelType.OpenSource,
-            "params": 32,
+            "organization": "QwQ Team",
             "text_score": 7.1,
             "visual_score": 8.964,
             "total_score": 16.064,
@@ -217,14 +215,13 @@ def get_leaderboard_df():
             "execution_success": 4.1,
             "file_generation": 1.5,
             "file_quality": 1.5,
-            "submission_date": "2025-02-05",
-            "license": "Apache 2.0"
+            "submission_date": "2025-02-05"
         },
         {
             "rank": 6,
             "model": "Qwen-2.5-32B",
             "model_type": ModelType.OpenSource,
-            "params": 32,
+            "organization": "Alibaba",
             "text_score": 0.7,
             "visual_score": 1.126,
             "total_score": 1.826,
@@ -233,8 +230,7 @@ def get_leaderboard_df():
             "execution_success": 0.5,
             "file_generation": 0.1,
             "file_quality": 0.1,
-            "submission_date": "2025-02-10",
-            "license": "Apache 2.0"
+            "submission_date": "2025-02-10"
         }
     ]
     
@@ -261,25 +257,73 @@ def init_leaderboard(dataframe):
             cant_deselect=[c.name for c in COLUMNS if c.never_hidden],
             label="Select Columns to Display:",
         ),
-        search_columns=["model", "license"],
+        search_columns=["model", "organization"],
         hide_columns=[c.name for c in COLUMNS if c.hidden],
         filter_columns=[
             ColumnFilter("model_type", type="checkboxgroup", label="Model types"),
-            # ColumnFilter(
-            #     "params",
-            #     type="slider",
-            #     min=0.01,
-            #     max=1500,
-            #     label="Select the number of parameters (B)",
-            # ),
+            ColumnFilter("organization", type="checkboxgroup", label="Organizations"),
         ],
         interactive=False,
     )
 
+# Function to handle ZIP file upload and extraction
+def process_zip_submission(zip_file):
+    if zip_file is None:
+        return "No file uploaded. Please upload a ZIP file containing your submission."
+    
+    # Create temp directory for extraction
+    temp_dir = tempfile.mkdtemp()
+    
+    try:
+        # Extract the zip file
+        with zipfile.ZipFile(zip_file.name, 'r') as zip_ref:
+            zip_ref.extractall(temp_dir)
+        
+        # Check for required files
+        model_info_path = os.path.join(temp_dir, "model_info.json")
+        results_json_path = os.path.join(temp_dir, "PhysCodeEval_results.json")
+        
+        if not os.path.exists(model_info_path):
+            return "Error: model_info.json not found in the ZIP file."
+        
+        if not os.path.exists(results_json_path):
+            return "Error: PhysCodeEval_results.json not found in the ZIP file."
+        
+        # Load model info
+        with open(model_info_path, 'r') as f:
+            model_info = json.load(f)
+        
+        # Check for required model info fields
+        required_fields = ["model_name", "model_type", "organization"]
+        missing_fields = [field for field in required_fields if field not in model_info]
+        
+        if missing_fields:
+            return f"Error: Missing required fields in model_info.json: {', '.join(missing_fields)}"
+        
+        # TODO: Process the submission files (this would involve your validation logic)
+        
+        return f"Successfully processed submission for {model_info['model_name']} by {model_info['organization']}. Your submission will be reviewed and added to the leaderboard once approved."
+    
+    except zipfile.BadZipFile:
+        return "Error: Invalid ZIP file."
+    except Exception as e:
+        return f"Error processing submission: {str(e)}"
+    finally:
+        # Clean up
+        shutil.rmtree(temp_dir)
+
 # Submission form handling
-def process_submission(model_name, model_type, license_type, submission_link):
+def process_submission(model_name, model_type, organization, team_name, email, submission_link):
+    # Check for required fields
+    if not model_name:
+        return "Error: Model name is required."
+    if not model_type:
+        return "Error: Model type is required."
+    if not email:
+        return "Error: Contact email is required."
+    
     # This would be implemented to handle actual submission processing
-    return f"Thank you for submitting {model_name}! Your submission will be reviewed and added to the leaderboard once verified."
+    return f"Thank you for submitting {model_name} from {organization or team_name}! Your submission will be reviewed and added to the leaderboard once verified. We will contact you at {email} if we need additional information."
 
 # Main application
 def create_demo():
@@ -322,27 +366,42 @@ def create_demo():
             with gr.TabItem("🚀 Submit", id=3):
                 gr.Markdown(SUBMISSION_TEXT, elem_classes="markdown-text")
                 
+                gr.Markdown("### Option 1: Upload Submission ZIP File")
+                with gr.Row():
+                    zip_file_input = gr.File(label="Upload submission ZIP file")
+                
+                zip_submit_button = gr.Button("Submit ZIP File")
+                zip_submission_result = gr.Markdown()
+                
+                zip_submit_button.click(
+                    process_zip_submission,
+                    [zip_file_input],
+                    zip_submission_result,
+                )
+                
+                gr.Markdown("### Option 2: Submit Form")
                 with gr.Row():
                     with gr.Column():
-                        model_name_input = gr.Textbox(label="Model Name")
+                        model_name_input = gr.Textbox(label="Model Name*")
                         model_type_input = gr.Dropdown(
-                            choices=["CloseSource", "Open Source", "API"],
-                            label="Model Type",
+                            choices=["Open Source", "Close Source", "API", "Proprietary"],
+                            label="Model Type*",
                             multiselect=False,
                         )
-                        #params_input = gr.Number(label="Parameters (billions)")
+                        organization_input = gr.Textbox(label="Organization (if applicable)")
                     
                     with gr.Column():
-                        license_input = gr.Textbox(label="License")
+                        team_name_input = gr.Textbox(label="Team Name (if applicable)")
+                        email_input = gr.Textbox(label="Contact Email*")
                         submission_link_input = gr.Textbox(label="GitHub Pull Request URL")
                 
-                submit_button = gr.Button("Submit")
-                submission_result = gr.Markdown()
+                form_submit_button = gr.Button("Submit Form")
+                form_submission_result = gr.Markdown()
                 
-                submit_button.click(
+                form_submit_button.click(
                     process_submission,
-                    [model_name_input, model_type_input, license_input, submission_link_input],
-                    submission_result,
+                    [model_name_input, model_type_input, organization_input, team_name_input, email_input, submission_link_input],
+                    form_submission_result,
                 )
         
         with gr.Row():