app.py

import gradio as gr
import pandas as pd
import os
import shutil

# Description and Introduction texts
DESCRIPTION = """
<h2 style='text-align: center; color: #cbff4d !important; text-shadow: 2px 2px 4px rgba(0,0,0,0.1);'>🚀 LLM Inference Leaderboard: Pushing the Boundaries of Performance 🚀</h2>
"""

INTRODUCTION = """
<div style='background-color: #e6ffd9; padding: 20px; border-radius: 15px; margin-bottom: 20px; box-shadow: 0 4px 6px rgba(0,0,0,0.1);'>
<h3 style='color: #00480a;'>🔬 Our Exciting Quest</h3>
<p style='color: #00480a;'>We're on a thrilling journey to help developers discover the perfect LLMs and libraries for their innovative projects! We've put these models through their paces using six cutting-edge inference engines:</p>
<ul style='color: #00480a;'>
    <li>🚄 vLLM</li>
    <li>🌟 TGI</li>
    <li>⚡ TensorRT-LLM</li>
    <li>🔮 Tritonvllm</li>
    <li>🚀 Deepspeed-mii</li>
    <li>🎯 ctranslate</li>
</ul>
<p style='color: #00480a;'>All our tests were conducted on state-of-the-art A100 GPUs hosted on Azure, ensuring a fair and neutral battleground!</p>
<p style='color: #00480a; font-weight: bold;'>Our mission: Empower developers, researchers, and AI enthusiasts to find their perfect LLM match for both development and production environments!</p>
</div>
"""

HOW_WE_TESTED = """
<div style='background-color: #cbff4d; padding: 20px; border-radius: 15px; margin-top: 20px; box-shadow: 0 4px 6px rgba(0,0,0,0.1);'>
<h3 style='color: #00480a;'>🧪 Our Rigorous Testing Process</h3>
<p style='color: #00480a;'>We left no stone unturned in our quest for reliable benchmarks:</p>
<ul style='color: #00480a;'>
    <li><strong>🖥️ Platform:</strong> A100 GPUs from Azure - the ultimate testing ground!</li>
    <li><strong>🐳 Setup:</strong> Docker containers for each library, ensuring a pristine environment.</li>
    <li><strong>⚙️ Configuration:</strong> Standardized settings (temperature 0.5, top_p 1) for laser-focused performance comparisons.</li>
    <li><strong>📊 Prompts & Token Ranges:</strong> Six diverse prompts, input lengths from 20 to 2,000 tokens, and generation lengths of 100, 200, and 500 tokens - pushing the boundaries of flexibility!</li>
    <li><strong>🤖 Models & Libraries Tested:</strong> We put the best through their paces: Phi-3-medium-128k-instruct, Meta-Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3, Qwen2-7B-Instruct, and Gemma-2-9b-it, using TGI, vLLM, DeepSpeed Mii, CTranslate2, Triton with vLLM Backend, and TensorRT-LLM.</li>
</ul>
</div>
"""

csv_folder_path = 'result_csv/'

UPLOAD_SECRET = os.getenv("UPLOAD_SECRET")

# ... [Previous functions remain the same: read_and_process_csv_files, get_leaderboard_df, filter_and_search] ...

def add_new_entry(file, password):
    global df
    if file is None:
        return df, "No file uploaded."
    
    if password != UPLOAD_SECRET:
        return df, "Incorrect password. Upload failed."
    
    new_df = pd.read_csv(file.name)
    
    columns_order = [
        "Model_Name", "Library", "TTFT", "Tokens-per-Second", "Token_Count", 
        "Input_Tokens", "Output_Tokens"
    ]
    for col in columns_order:
        if col not in new_df.columns:
            new_df[col] = pd.NA
    new_df = new_df[columns_order]
    
    df = pd.concat([df, new_df], ignore_index=True)
    
    filename = os.path.basename(file.name)
    destination = os.path.join(csv_folder_path, filename)
    shutil.copy(file.name, destination)
    
    return df, f"File '{filename}' uploaded and data added successfully!"

def read_and_process_csv_files(folder_path):
    all_data = []
    for filename in os.listdir(folder_path):
        if filename.endswith('.csv'):
            file_path = os.path.join(folder_path, filename)
            df = pd.read_csv(file_path)
            all_data.append(df)
    
    combined_df = pd.concat(all_data, ignore_index=True)
    
    columns_order = [
        "Model_Name", "Library", "TTFT", "Tokens-per-Second", "Token_Count", 
        "input_length", "output_length"
    ]
    
    for col in columns_order:
        if col not in combined_df.columns:
            combined_df[col] = pd.NA
    
    return combined_df[columns_order]

df = read_and_process_csv_files(csv_folder_path)

def get_leaderboard_df():
    return df

def filter_and_search(search_term, library_filter):
    filtered_df = df.copy()
    
    if search_term:
        filtered_df = filtered_df[filtered_df['Model_Name'].str.contains(search_term, case=False, na=False)]
    
    if library_filter != "All":
        filtered_df = filtered_df[filtered_df['Library'] == library_filter]
    
    return filtered_df

custom_css = """
body {
    background-color: #f0fff0;
    font-family: 'Roboto', sans-serif;
}
.gradio-container {
    max-width: 1200px !important;
}
.gradio-container .prose * {
    color: #00480a !important;
}
.gradio-container .prose h2,
.gradio-container .prose h3 {
    color: #00480a !important;
}
.tabs {
    background-color: #e6ffd9;
    border-radius: 15px;
    overflow: hidden;
    box-shadow: 0 4px 6px rgba(0,0,0,0.1);
}
.tab-nav {
    background-color: #00480a;
    padding: 10px;
}
.tab-nav button {
    color: #cbff4d !important;
    background-color: #006400;
    border: none;
    padding: 10px 20px;
    margin-right: 5px;
    border-radius: 10px;
    cursor: pointer;
    transition: all 0.3s ease;
}
.tab-nav button:hover {
    background-color: #cbff4d;
    color: #00480a !important;
}
.tab-nav button.selected {
    background-color: #cbff4d;
    color: #00480a !important;
    font-weight: bold;
}
.gr-button-primary {
    background-color: #00480a !important;
    border-color: #00480a !important;
    color: #cbff4d !important;
}
.gr-button-primary:hover {
    background-color: #cbff4d !important;
    color: #00480a !important;
}
"""

with gr.Blocks(css=custom_css) as demo:
    gr.HTML(DESCRIPTION)
    gr.HTML(INTRODUCTION)
    
    with gr.Tabs():
        with gr.TabItem("📊 Leaderboard"):
            with gr.Row():
                search_input = gr.Textbox(label="🔍 Search Model Name", placeholder="Enter model name...")
                library_dropdown = gr.Dropdown(choices=["All"] + df['Library'].unique().tolist(), label="🏷️ Filter by Library", value="All")
            
            leaderboard = gr.DataFrame(df)
            
            gr.HTML(HOW_WE_TESTED)
        
        with gr.TabItem("➕ Add New Entry"):
            file_upload = gr.File(label="📁 Upload CSV File")
            password_input = gr.Textbox(label="🔐 Upload Password", type="password")
            submit_button = gr.Button("📤 Add Entry", variant="primary")
            result = gr.Markdown()

    submit_button.click(
        add_new_entry,
        inputs=[file_upload, password_input],
        outputs=[leaderboard, result]
    )

    search_input.change(filter_and_search, inputs=[search_input, library_dropdown], outputs=leaderboard)
    library_dropdown.change(filter_and_search, inputs=[search_input, library_dropdown], outputs=leaderboard)

    demo.load(get_leaderboard_df, outputs=[leaderboard])

if __name__ == "__main__":
    demo.launch()