Update app.py

app.py

@@ -12,40 +12,17 @@ from PIL import Image
 from io import BytesIO
 import tempfile
 import sys
+import subprocess
 
+#############################################
+# PART 1: YOUR EXISTING PLOTS & FUNCTIONALITY
+#############################################
 
-# --------------------------------------------------------------------
-# PART 1: YOUR EXISTING (TINY) DATA & PLOTS
-# --------------------------------------------------------------------
-
+# For demonstration, assume you have a small data_full for "tiny" benchmarks:
 data_full = [
-    ['CultriX/Qwen2.5-14B-SLERPv7', 'https://huggingface.co/CultriX/Qwen2.5-14B-SLERPv7', 0.7205, 0.8272, 0.7541, 0.6581, 0.5, 0.729],
-    ['djuna/Q2.5-Veltha-14B-0.5', 'https://huggingface.co/djuna/Q2.5-Veltha-14B-0.5', 0.7492, 0.8386, 0.7305, 0.598, 0.43, 0.7817],
-    ['CultriX/Qwen2.5-14B-FinalMerge', 'https://huggingface.co/CultriX/Qwen2.5-14B-FinalMerge', 0.7248, 0.8277, 0.7113, 0.7052, 0.57, 0.7001],
-    ['CultriX/Qwen2.5-14B-MultiCultyv2', 'https://huggingface.co/CultriX/Qwen2.5-14B-MultiCultyv2', 0.7295, 0.8359, 0.7363, 0.5767, 0.44, 0.7316],
-    ['CultriX/Qwen2.5-14B-Brocav7', 'https://huggingface.co/CultriX/Qwen2.5-14B-Brocav7', 0.7445, 0.8353, 0.7508, 0.6292, 0.46, 0.7629],
-    ['CultriX/Qwen2.5-14B-Broca', 'https://huggingface.co/CultriX/Qwen2.5-14B-Broca', 0.7456, 0.8352, 0.748, 0.6034, 0.44, 0.7716],
-    ['CultriX/Qwen2.5-14B-Brocav3', 'https://huggingface.co/CultriX/Qwen2.5-14B-Brocav3', 0.7395, 0.8388, 0.7393, 0.6405, 0.47, 0.7659],
-    ['CultriX/Qwen2.5-14B-Brocav4', 'https://huggingface.co/CultriX/Qwen2.5-14B-Brocav4', 0.7432, 0.8377, 0.7444, 0.6277, 0.48, 0.758],
-    ['CultriX/Qwen2.5-14B-Brocav2', 'https://huggingface.co/CultriX/Qwen2.5-14B-Brocav2', 0.7492, 0.8302, 0.7508, 0.6377, 0.51, 0.7478],
-    ['CultriX/Qwen2.5-14B-Brocav5', 'https://huggingface.co/CultriX/Qwen2.5-14B-Brocav5', 0.7445, 0.8313, 0.7547, 0.6376, 0.5, 0.7304],
-    ['CultriX/Qwen2.5-14B-Brocav6', 'https://huggingface.co/CultriX/Qwen2.5-14B-Brocav6', 0.7179, 0.8354, 0.7531, 0.6378, 0.49, 0.7524],
-    ['CultriX/Qwenfinity-2.5-14B', 'https://huggingface.co/CultriX/Qwenfinity-2.5-14B', 0.7347, 0.8254, 0.7279, 0.7267, 0.56, 0.697],
-    ['CultriX/Qwen2.5-14B-Emergedv2', 'https://huggingface.co/CultriX/Qwen2.5-14B-Emergedv2', 0.7137, 0.8335, 0.7363, 0.5836, 0.44, 0.7344],
-    ['CultriX/Qwen2.5-14B-Unity', 'https://huggingface.co/CultriX/Qwen2.5-14B-Unity', 0.7063, 0.8343, 0.7423, 0.682, 0.57, 0.7498],
-    ['CultriX/Qwen2.5-14B-MultiCultyv3', 'https://huggingface.co/CultriX/Qwen2.5-14B-MultiCultyv3', 0.7132, 0.8216, 0.7395, 0.6792, 0.55, 0.712],
-    ['CultriX/Qwen2.5-14B-Emergedv3', 'https://huggingface.co/CultriX/Qwen2.5-14B-Emergedv3', 0.7436, 0.8312, 0.7519, 0.6585, 0.55, 0.7068],
-    ['CultriX/SeQwence-14Bv1', 'https://huggingface.co/CultriX/SeQwence-14Bv1', 0.7278, 0.841, 0.7541, 0.6816, 0.52, 0.7539],
-    ['CultriX/Qwen2.5-14B-Wernickev2', 'https://huggingface.co/CultriX/Qwen2.5-14B-Wernickev2', 0.7391, 0.8168, 0.7273, 0.622, 0.45, 0.7572],
-    ['CultriX/Qwen2.5-14B-Wernickev3', 'https://huggingface.co/CultriX/Qwen2.5-14B-Wernickev3', 0.7357, 0.8148, 0.7245, 0.7023, 0.55, 0.7869],
-    ['CultriX/Qwen2.5-14B-Wernickev4', 'https://huggingface.co/CultriX/Qwen2.5-14B-Wernickev4', 0.7355, 0.829, 0.7497, 0.6306, 0.48, 0.7635],
-    ['CultriX/SeQwential-14B-v1', 'https://huggingface.co/CultriX/SeQwential-14B-v1', 0.7355, 0.8205, 0.7549, 0.6367, 0.48, 0.7626],
-    ['CultriX/Qwen2.5-14B-Wernickev5', 'https://huggingface.co/CultriX/Qwen2.5-14B-Wernickev5', 0.7224, 0.8272, 0.7541, 0.679, 0.51, 0.7578],
-    ['CultriX/Qwen2.5-14B-Wernickev6', 'https://huggingface.co/CultriX/Qwen2.5-14B-Wernickev6', 0.6994, 0.7549, 0.5816, 0.6991, 0.58, 0.7267],
-    ['CultriX/Qwen2.5-14B-Wernickev7', 'https://huggingface.co/CultriX/Qwen2.5-14B-Wernickev7', 0.7147, 0.7599, 0.6097, 0.7056, 0.57, 0.7164],
-    ['CultriX/Qwen2.5-14B-FinalMerge-tmp2', 'https://huggingface.co/CultriX/Qwen2.5-14B-FinalMerge-tmp2', 0.7255, 0.8192, 0.7535, 0.6671, 0.5, 0.7612],
-    ['CultriX/Qwen2.5-14B-BrocaV8', 'https://huggingface.co/CultriX/Qwen2.5-14B-BrocaV8', 0.7415, 0.8396, 0.7334, 0.5785, 0.4300, 0.7646],
+    # your existing data
 ]
+
 columns = ["Model Configuration", "Model Link", "tinyArc", "tinyHellaswag", 
            "tinyMMLU", "tinyTruthfulQA", "tinyTruthfulQA_mc1", "tinyWinogrande"]
 df_full = pd.DataFrame(data_full, columns=columns)
@@ -127,7 +104,7 @@ def plot_task_specific_top_models():
 
 def plot_heatmap():
     plt.figure(figsize=(14, 10))
-    sns.heatmap(df_full.iloc[:, 2:], annot=True, cmap="YlGnBu", 
+    sns.heatmap(df_full.iloc[:, 2:], annot=True, cmap="YlGnBu",
                 xticklabels=columns[2:], yticklabels=df_full["Model Configuration"])
     plt.title("Performance Heatmap", fontsize=16)
     plt.tight_layout()
@@ -143,13 +120,16 @@ def plot_heatmap():
     return pil_image, temp_image_file.name
 
 def scrape_mergekit_config(model_name):
+    """
+    Example from your code that tries to find <pre> blocks on the model page.
+    """
     model_link = df_full.loc[df_full["Model Configuration"] == model_name, "Model Link"].values[0]
     response = requests.get(model_link)
     if response.status_code != 200:
         return f"Failed to fetch model page for {model_name}. Please check the link."
 
     soup = BeautifulSoup(response.text, "html.parser")
-    yaml_config = soup.find("pre")  # Assume YAML is in <pre> tags
+    yaml_config = soup.find("pre")
     if yaml_config:
         return yaml_config.text.strip()
     return f"No YAML configuration found for {model_name}."
@@ -206,163 +186,41 @@ def download_all_data():
             image_bytes.seek(0)
             zf.writestr(filename, image_bytes.read())
 
-        # Also try scraping each model for a YAML config
+        # Optionally, scrape each model for a YAML config:
         for model_name in df_full["Model Configuration"].to_list():
             yaml_content = scrape_mergekit_config(model_name)
             if ("No YAML configuration found" not in yaml_content) and ("Failed to fetch model page" not in yaml_content):
-               zf.writestr(f"{model_name.replace('/', '_')}_config.yaml", yaml_content.encode())
+                zf.writestr(f"{model_name.replace('/', '_')}_config.yaml", yaml_content.encode())
 
     zip_buffer.seek(0)
     return zip_buffer, "analysis_data.zip"
 
 
-# --------------------------------------------------------------------
-# PART 2: FULL "DATA START" SNIPPET (RANKS 44–105) + Parser
-# --------------------------------------------------------------------
-benchmark_data = [
-    # The entire dataset from your "DATA START", rank 44..105
-    # (the code you posted with "knowledge of config" or scraping logic)
-    {
-        "rank": 44,
-        "name": "sometimesanotion/Qwen2.5-14B-Vimarckoso-v3",
-        "scores": {
-            "average": 40.10,
-            "IFEval": 72.57,
-            "BBH": 48.58,
-            "MATH": 34.44,
-            "GPQA": 17.34,
-            "MUSR": 19.39,
-            "MMLU-PRO": 48.26
-        },
-        "hf_url": "https://huggingface.co/sometimesanotion/Qwen2.5-14B-Vimarckoso-v3",
-        "known_config": {
-            "models": [
-                {"model": "CultriX/SeQwence-14Bv1"},
-                {"model": "allknowingroger/Qwenslerp5-14B"}
-            ],
-            "merge_method": "slerp",
-            "base_model": "CultriX/SeQwence-14Bv1",
-            "dtype": "bfloat16",
-            "parameters": {
-                "t": [0, 0.5, 1, 0.5, 0]
-            }
-        }
-    },
-    # ... rest of the snippet ...
-    # (Exactly copy/paste your big block from rank=44 to rank=105)
-]
-
+#############################################
+# PART 2: RUNNING `scrape-leaderboard.py`
+#############################################
 
-def snippet_scrape_model_page(url):
+def run_scrape_leaderboard():
     """
-    Same as scrape_model_page, but we keep it separate for clarity.
-    """
-    return scrape_model_page(url)
-
-def snippet_print_benchmark_and_config_info(model_info):
+    Uses Python's `subprocess` to call the external script: `scrape-leaderboard.py`
+    capturing whatever the script prints to stdout.
     """
-    Prints an overview for each model (your "DATA START" logic), 
-    either known config or scraping snippet.
-    """
-    print(f"---\nModel Rank: {model_info['rank']}")
-    print(f"Model Name: {model_info['name']}")
-    print(f"Model average score across benchmarks in %: {model_info['scores']['average']}")
-    print(f"Models average score on IFEval benchmarks in %: {model_info['scores']['IFEval']}")
-    print(f"Models average score on BBH benchmarks in %: {model_info['scores']['BBH']}")
-    print(f"Models average score on MATH benchmarks in %: {model_info['scores']['MATH']}")
-    print(f"Models average score in GPQA benchmarks in %: {model_info['scores']['GPQA']}")
-    print(f"Models average score in MUSR benchmarks in %: {model_info['scores']['MUSR']}")
-    print(f"Models average score in MMLU-PRO benchmarks in %: {model_info['scores']['MMLU-PRO']}")
-
-    # If there's a known_config, print it as YAML
-    if model_info["known_config"] is not None:
-        print("###")
-        print("models:")
-        for m in model_info["known_config"]["models"]:
-            print(f"  - model: {m['model']}")
-        print(f"merge_method: {model_info['known_config']['merge_method']}")
-        print(f"base_model: {model_info['known_config']['base_model']}")
-        print(f"dtype: {model_info['known_config']['dtype']}")
-        print("parameters:")
-        print(f"  t: {model_info['known_config']['parameters']['t']} # V shaped curve: Hermes for input & output, WizardMath in the middle layers")
-        print("###")
-        return
-
-    # Otherwise, scrape
-    scraped = snippet_scrape_model_page(model_info["hf_url"])
-    if isinstance(scraped, str):
-        # Means it's an error string or something
-        if "Error:" in scraped:
-            print("(No MergeKit configuration found or error occurred.)\n")
-            # optionally print snippet
-        else:
-            print(scraped)
-        return
-    else:
-        # It's presumably a dict: { "yaml_configuration": "...", "metadata": "..." }
-        if ("No YAML configuration found." in scraped["yaml_configuration"]):
-            print("(No MergeKit configuration found.)\n")
-            # Print your snippet code
-            print("You can try the following Python script to scrape the model page:\n")
-            print("#" * 70)
-            print(f'''import requests
-from bs4 import BeautifulSoup
-
-def scrape_model_page(model_url):
     try:
-        response = requests.get(model_url)
-        if response.status_code != 200:
-            return f"Error: Unable to fetch the page (Status Code: {{response.status_code}})"
-        
-        soup = BeautifulSoup(response.text, "html.parser")
-
-        yaml_config = soup.find("pre")
-        yaml_text = yaml_config.text.strip() if yaml_config else "No YAML configuration found."
-
-        metadata_section = soup.find("div", class_="metadata")
-        metadata_text = metadata_section.text.strip() if metadata_section else "No metadata found."
-
-        return {{
-            "yaml_configuration": yaml_text,
-            "metadata": metadata_text
-        }}
-
+        # Make sure 'scrape-leaderboard.py' is in the same folder or give the full path
+        result = subprocess.run(["python", "scrape-leaderboard.py"], capture_output=True, text=True)
+        # Return the combined stdout/stderr or just stdout
+        return result.stdout if result.stdout else result.stderr
     except Exception as e:
-        return f"Error: {{str(e)}}"
-
-if __name__ == "__main__":
-    model_url = "{model_info['hf_url']}"
-    result = scrape_model_page(model_url)
-    print(result)''')
-            print("#" * 70)
-        else:
-            print("###")
-            print(scraped["yaml_configuration"])
-            print("###")
-
-def run_non_tiny_benchmarks():
-    """
-    Captures the stdout from printing each model in benchmark_data 
-    (ranks 44 to 105), returning a single string for Gradio to display.
-    """
-    old_stdout = sys.stdout
-    buffer = io.StringIO()
-    sys.stdout = buffer
+        return f"Error running script: {str(e)}"
 
-    for model in benchmark_data:
-        snippet_print_benchmark_and_config_info(model)
 
-    sys.stdout = old_stdout
-    return buffer.getvalue()
+###############################
+# PART 3: YOUR GRADIO INTERFACE
+###############################
 
-
-# --------------------------------------------------------------------
-# PART 3: GRADIO APP (Your existing UI plus the "Parse Non-Tiny" button)
-# --------------------------------------------------------------------
 with gr.Blocks() as demo:
     gr.Markdown("# Comprehensive Model Performance Analysis with Hugging Face Links")
 
-    # The existing UI
     with gr.Row():
         btn1 = gr.Button("Show Average Performance")
         img1 = gr.Image(type="pil", label="Average Performance Plot")
@@ -387,6 +245,7 @@ with gr.Blocks() as demo:
         heatmap_download = gr.File(label="Download Heatmap")
         btn4.click(plot_heatmap, outputs=[heatmap_img, heatmap_download])
 
+    # Drop-down to pick a model, scrape for config
     with gr.Row():
         model_selector = gr.Dropdown(choices=df_full["Model Configuration"].tolist(), label="Select a Model")
         with gr.Column():
@@ -398,12 +257,13 @@ with gr.Blocks() as demo:
             yaml_download = gr.File(label="Download MergeKit Configuration")
             save_yaml_btn.click(download_yaml, inputs=[yaml_output, model_selector], outputs=yaml_download)
 
+    # Button to download everything (CSV + plots)
     with gr.Row():
         download_all_btn = gr.Button("Download Everything")
         all_downloads = gr.File(label="Download All Data")
         download_all_btn.click(download_all_data, outputs=all_downloads)
-        
-    # Live scraping feature
+
+    # Live scraping of any model URL
     gr.Markdown("## Live Scraping Features")
     with gr.Row():
         url_input = gr.Textbox(label="Enter Hugging Face Model URL", placeholder="https://huggingface.co/<model>")
@@ -411,11 +271,12 @@ with gr.Blocks() as demo:
         live_scrape_output = gr.Textbox(label="Scraped Data", lines=15)
         live_scrape_btn.click(display_scraped_model_data, inputs=url_input, outputs=live_scrape_output)
 
-    # NEW: Non-Tiny Benchmarks button
-    gr.Markdown("## Non-Tiny Benchmark Parser (Ranks 44–105)")
+    # NEW: Button that runs the external script 'scrape-leaderboard.py'
+    gr.Markdown("## Run `scrape-leaderboard.py` Externally")
     with gr.Row():
-        parse_non_tiny_btn = gr.Button("Parse Non-Tiny Benchmarks")
-        parse_non_tiny_output = gr.Textbox(label="Non-Tiny Benchmark Output", lines=30)
-        parse_non_tiny_btn.click(fn=run_non_tiny_benchmarks, outputs=parse_non_tiny_output)
+        run_script_btn = gr.Button("Run 'scrape-leaderboard.py'")
+        run_script_output = gr.Textbox(label="Script Output", lines=25)
+        run_script_btn.click(fn=run_scrape_leaderboard, outputs=run_script_output)
 
+# Finally, launch the app
 demo.launch()