utils.py

import pandas as pd
import gradio as gr
import csv
import json
import os
import requests
import io
import shutil
from huggingface_hub import Repository

HF_TOKEN = os.environ.get("HF_TOKEN")

BASE_COLS = ["Rank", "Models", "Model Size(B)", "Data Source"]
TASKS_V1 = ["V1-Overall", "I-CLS", "I-QA", "I-RET", "I-VG"]
TASKS_V2 = ["V2-Overall", "V-CLS", "V-QA", "V-RET", "V-MRET", "VisDoc"]
COLUMN_NAMES = BASE_COLS + TASKS_V1 + TASKS_V2

DATA_TITLE_TYPE = ['number', 'markdown', 'str', 'markdown'] + \
                    ['number'] * (len(TASKS_V1) + len(TASKS_V2))

LEADERBOARD_INTRODUCTION = """
# MMEB Leaderboard

## Introduction
We introduce a novel benchmark, MMEB (Massive Multimodal Embedding Benchmark), 
which includes 36 datasets spanning four meta-task categories: classification, visual question answering, retrieval, and visual grounding. MMEB provides a comprehensive framework for training
and evaluating embedding models across various combinations of text and image modalities. 
All tasks are reformulated as ranking tasks, where the model follows instructions, processes a query, and selects the correct target from a set of candidates. The query and target can be an image, text,
or a combination of both. MMEB is divided into 20 in-distribution datasets, which can be used for
training, and 16 out-of-distribution datasets, reserved for evaluation.

Building upon on **MMEB**, **MMEB-V2** expands the evaluation scope to include five new tasks: four video-based tasks 
— Video Retrieval, Moment Retrieval, Video Classification, and Video Question Answering — and one task focused on visual documents, Visual Document Retrieval. 
This comprehensive suite enables robust evaluation of multimodal embedding models across static, temporal, and structured visual data settings.

| [**Overview**](https://tiger-ai-lab.github.io/VLM2Vec/) | [**Github**](https://github.com/TIGER-AI-Lab/VLM2Vec) 
| [**📖MMEB-V2/VLM2Vec-V2 Paper (TBA)**](https://arxiv.org/abs/2410.05160) 
| [**📖MMEB-V1/VLM2Vec-V1 Paper**](https://arxiv.org/abs/2410.05160) 
| [**Hugging Face**](https://huggingface.co/datasets/TIGER-Lab/MMEB-V2)
"""

TABLE_INTRODUCTION = """"""

LEADERBOARD_INFO = """
## Dataset Summary
"""

CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
CITATION_BUTTON_TEXT = r"""@article{jiang2024vlm2vec,
  title={VLM2Vec: Training Vision-Language Models for Massive Multimodal Embedding Tasks},
  author={Jiang, Ziyan and Meng, Rui and Yang, Xinyi and Yavuz, Semih and Zhou, Yingbo and Chen, Wenhu},
  journal={arXiv preprint arXiv:2410.05160},
  year={2024}
}"""

SUBMIT_INTRODUCTION = """# Submit on MMEB Leaderboard Introduction

## ⚠ Please note that you need to submit the JSON file with the following format:

### **TO SUBMIT V1 ONLY**
```json
[
    {
        "Model": "<Model Name>",
        <Optional>"URL": "<Model URL>",
        "Model Size(B)": 1000,
        "Data Source": Self-Reported,
        "V1-Overall": 50.0,
        "I-CLS": 50.0,
        "I-QA": 50.0,
        "I-RET": 50.0,
        "I-VG": 50.0
    }, 
]
```

### **TO SUBMIT V2 ONLY**
```json
[
    {
        "Model": "<Model Name>",
        <Optional>"URL": "<Model URL>",
        "Model Size(B)": 1000,
        "Data Source": Self-Reported,
        "V2-Overall": 50.0,
        "V-CLS": 50.0,
        "V-QA": 50.0,
        "V-RET": 50.0,
        "V-MRET": 50.0, 
        "VisDoc": 50.0
    }, 
]
```
You are also welcome to submit both versions by including all the fields above! :) \n
You may refer to the [**GitHub page**](https://github.com/TIGER-AI-Lab/VLM2Vec) for instructions about evaluating your model. \n
Please send us an email at m7su@uwaterloo.ca, attaching the JSON file. We will review your submission and update the leaderboard accordingly.
"""

def create_hyperlinked_names(df):
    def convert_url(url, model_name):
        return f'<a href="{url}">{model_name}</a>' if url is not None else model_name

    def add_link_to_model_name(row):
        row['Models'] = convert_url(row['URL'], row['Models'])
        return row
    
    df = df.copy()
    df = df.apply(add_link_to_model_name, axis=1)
    return df

# def fetch_data(file: str) -> pd.DataFrame:
#     # fetch the leaderboard data from remote
#     if file is None:
#         raise ValueError("URL Not Provided")
#     url = f"https://huggingface.co/spaces/TIGER-Lab/MMEB/resolve/main/{file}"
#     print(f"Fetching data from {url}")
#     response = requests.get(url)
#     if response.status_code != 200:
#         raise requests.HTTPError(f"Failed to fetch data: HTTP status code {response.status_code}")
#     return pd.read_json(io.StringIO(response.text), orient='records', lines=True)

def get_df(file="results.jsonl"):
    df = pd.read_json(file, orient='records', lines=True)
    df['Model Size(B)'] = df['Model Size(B)'].apply(process_model_size)
    df = df.sort_values(by=['V1-Overall'], ascending=False)
    df = create_hyperlinked_names(df)
    df['Rank'] = range(1, len(df) + 1)
    return df

def refresh_data():
    df = get_df()
    return df[COLUMN_NAMES]

def search_and_filter_models(df, query, min_size, max_size):
    filtered_df = df.copy()
    
    if query:
        filtered_df = filtered_df[filtered_df['Models'].str.contains(query, case=False, na=False)]

    size_mask = filtered_df['Model Size(B)'].apply(lambda x: 
        (min_size <= 1000.0 <= max_size) if x == 'unknown' 
        else (min_size <= x <= max_size))
    
    filtered_df = filtered_df[size_mask]
    
    return filtered_df[COLUMN_NAMES]


def search_models(df, query):
    if query:
        return df[df['Models'].str.contains(query, case=False, na=False)]
    return df

def get_size_range(df):
    sizes = df['Model Size(B)'].apply(lambda x: 0.0 if x == 'unknown' else x)
    if (sizes == 0.0).all():
        return 0.0, 1000.0
    return float(sizes.min()), float(sizes.max())


def process_model_size(size):
    if pd.isna(size) or size == 'unk':
        return 'unknown'
    try:
        val = float(size)
        return val
    except (ValueError, TypeError):
        return 'unknown'


def filter_columns_by_tasks(df, selected_tasks=None):
    if selected_tasks is None or len(selected_tasks) == 0:
        return df[COLUMN_NAMES]
    
    base_columns = ['Models', 'Model Size(B)', 'Data Source', 'Overall']
    selected_columns = base_columns + selected_tasks
    
    available_columns = [col for col in selected_columns if col in df.columns]
    return df[available_columns]