update leaderboard

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 scale-hf-logo.png filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,13 +1,13 @@
 ---
-title: VideoEval Pro
+title: VideoEval-Pro Leaderboard
 emoji: 🥇
 colorFrom: green
 colorTo: indigo
 sdk: gradio
 app_file: app.py
-pinned: true
+pinned: false
 license: apache-2.0
-short_description: VideoEval-Pro Leaderboard
+short_description: A more robust benchmark for long video understanding.
 sdk_version: 5.19.0
 ---
 

app.py

@@ -1,204 +1,125 @@
-import gradio as gr
-from gradio_leaderboard import Leaderboard, ColumnFilter, SelectColumns
-import pandas as pd
-from apscheduler.schedulers.background import BackgroundScheduler
-from huggingface_hub import snapshot_download
-
-from src.about import (
-    CITATION_BUTTON_LABEL,
-    CITATION_BUTTON_TEXT,
-    EVALUATION_QUEUE_TEXT,
-    INTRODUCTION_TEXT,
-    LLM_BENCHMARKS_TEXT,
-    TITLE,
-)
-from src.display.css_html_js import custom_css
-from src.display.utils import (
-    BENCHMARK_COLS,
-    COLS,
-    EVAL_COLS,
-    EVAL_TYPES,
-    AutoEvalColumn,
-    ModelType,
-    fields,
-    WeightType,
-    Precision
-)
-from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN
-from src.populate import get_evaluation_queue_df, get_leaderboard_df
-from src.submission.submit import add_new_eval
-
-
-def restart_space():
-    API.restart_space(repo_id=REPO_ID)
-
-### Space initialisation
-try:
-    print(EVAL_REQUESTS_PATH)
-    snapshot_download(
-        repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
-    )
-except Exception:
-    restart_space()
-try:
-    print(EVAL_RESULTS_PATH)
-    snapshot_download(
-        repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
-    )
-except Exception:
-    restart_space()
-
-
-LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)
-
-(
-    finished_eval_queue_df,
-    running_eval_queue_df,
-    pending_eval_queue_df,
-) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
-
-def init_leaderboard(dataframe):
-    if dataframe is None or dataframe.empty:
-        raise ValueError("Leaderboard DataFrame is empty or None.")
-    return Leaderboard(
-        value=dataframe,
-        datatype=[c.type for c in fields(AutoEvalColumn)],
-        select_columns=SelectColumns(
-            default_selection=[c.name for c in fields(AutoEvalColumn) if c.displayed_by_default],
-            cant_deselect=[c.name for c in fields(AutoEvalColumn) if c.never_hidden],
-            label="Select Columns to Display:",
-        ),
-        search_columns=[AutoEvalColumn.model.name, AutoEvalColumn.license.name],
-        hide_columns=[c.name for c in fields(AutoEvalColumn) if c.hidden],
-        filter_columns=[
-            ColumnFilter(AutoEvalColumn.model_type.name, type="checkboxgroup", label="Model types"),
-            ColumnFilter(AutoEvalColumn.precision.name, type="checkboxgroup", label="Precision"),
-            ColumnFilter(
-                AutoEvalColumn.params.name,
-                type="slider",
-                min=0.01,
-                max=150,
-                label="Select the number of parameters (B)",
-            ),
-            ColumnFilter(
-                AutoEvalColumn.still_on_hub.name, type="boolean", label="Deleted/incomplete", default=True
-            ),
-        ],
-        bool_checkboxgroup_label="Hide models",
-        interactive=False,
-    )
-
-
-demo = gr.Blocks(css=custom_css)
-with demo:
-    gr.HTML(TITLE)
-    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
-
+from utils import *
+
+global data_component
+
+def update_table(query, min_size, max_size, selected_tasks=None, selected_type="All"):
+    df = get_df()
+    if selected_type and selected_type != "All":
+        df = df[df["Type"] == selected_type]
+    filtered_df = search_and_filter_models(df, query, min_size, max_size)
+    if selected_tasks and len(selected_tasks) > 0:
+        selected_columns = BASE_COLS + selected_tasks
+        filtered_df = filtered_df[selected_columns]
+    return filtered_df
+
+with gr.Blocks() as block:
+    gr.Markdown(LEADERBOARD_INTRODUCTION)
+    
     with gr.Tabs(elem_classes="tab-buttons") as tabs:
-        with gr.TabItem("🏅 LLM Benchmark", elem_id="llm-benchmark-tab-table", id=0):
-            leaderboard = init_leaderboard(LEADERBOARD_DF)
-
-        with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2):
-            gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
-
-        with gr.TabItem("🚀 Submit here! ", elem_id="llm-benchmark-tab-table", id=3):
-            with gr.Column():
-                with gr.Row():
-                    gr.Markdown(EVALUATION_QUEUE_TEXT, elem_classes="markdown-text")
-
-                with gr.Column():
-                    with gr.Accordion(
-                        f"✅ Finished Evaluations ({len(finished_eval_queue_df)})",
-                        open=False,
-                    ):
-                        with gr.Row():
-                            finished_eval_table = gr.components.Dataframe(
-                                value=finished_eval_queue_df,
-                                headers=EVAL_COLS,
-                                datatype=EVAL_TYPES,
-                                row_count=5,
-                            )
-                    with gr.Accordion(
-                        f"🔄 Running Evaluation Queue ({len(running_eval_queue_df)})",
-                        open=False,
-                    ):
-                        with gr.Row():
-                            running_eval_table = gr.components.Dataframe(
-                                value=running_eval_queue_df,
-                                headers=EVAL_COLS,
-                                datatype=EVAL_TYPES,
-                                row_count=5,
-                            )
-
-                    with gr.Accordion(
-                        f"⏳ Pending Evaluation Queue ({len(pending_eval_queue_df)})",
-                        open=False,
-                    ):
-                        with gr.Row():
-                            pending_eval_table = gr.components.Dataframe(
-                                value=pending_eval_queue_df,
-                                headers=EVAL_COLS,
-                                datatype=EVAL_TYPES,
-                                row_count=5,
-                            )
+        # Table 1
+        with gr.TabItem("📊 VideoEval-Pro", elem_id="qa-tab-table1", id=1):
             with gr.Row():
-                gr.Markdown("# ✉️✨ Submit your model here!", elem_classes="markdown-text")
+                with gr.Accordion("Citation", open=False):
+                    citation_button = gr.Textbox(
+                        value=CITATION_BUTTON_TEXT,
+                        label=CITATION_BUTTON_LABEL,
+                        elem_id="citation-button",
+                        lines=10,
+                    )
 
             with gr.Row():
-                with gr.Column():
-                    model_name_textbox = gr.Textbox(label="Model name")
-                    revision_name_textbox = gr.Textbox(label="Revision commit", placeholder="main")
-                    model_type = gr.Dropdown(
-                        choices=[t.to_str(" : ") for t in ModelType if t != ModelType.Unknown],
-                        label="Model type",
-                        multiselect=False,
-                        value=None,
-                        interactive=True,
-                    )
+                search_bar = gr.Textbox(
+                    placeholder="Search models...",
+                    show_label=False,
+                    elem_id="search-bar"
+                )
+            
+            df = get_df()
+            min_size, max_size = get_size_range(df)
 
-                with gr.Column():
-                    precision = gr.Dropdown(
-                        choices=[i.value.name for i in Precision if i != Precision.Unknown],
-                        label="Precision",
-                        multiselect=False,
-                        value="float16",
-                        interactive=True,
-                    )
-                    weight_type = gr.Dropdown(
-                        choices=[i.value.name for i in WeightType],
-                        label="Weights type",
-                        multiselect=False,
-                        value="Original",
-                        interactive=True,
-                    )
-                    base_model_name_textbox = gr.Textbox(label="Base model (for delta or adapter weights)")
+            with gr.Row():
+                min_size_slider = gr.Slider(
+                    minimum=min_size,
+                    maximum=max_size,
+                    value=min_size,
+                    step=0.1,
+                    label="Minimum number of parameters (B)",
+                )
+                max_size_slider = gr.Slider(
+                    minimum=min_size,
+                    maximum=max_size,
+                    value=max_size,
+                    step=0.1,
+                    label="Maximum number of parameters (B)",
+                )
+            
+            with gr.Row():
+                type_select = gr.Dropdown(
+                    choices=["All", "Proprietary", "Open-source"],
+                    value="All",
+                    label="Select Model Type",
+                    elem_id="type-select"
+                )
 
-            submit_button = gr.Button("Submit Eval")
-            submission_result = gr.Markdown()
-            submit_button.click(
-                add_new_eval,
-                [
-                    model_name_textbox,
-                    base_model_name_textbox,
-                    revision_name_textbox,
-                    precision,
-                    weight_type,
-                    model_type,
-                ],
-                submission_result,
+            with gr.Row():
+                tasks_select = gr.CheckboxGroup(
+                    choices=TASKS,
+                    value=OPEN_TASKS,
+                    label="Select tasks to Display",
+                    elem_id="tasks-select"
+                )
+
+            data_component = gr.components.Dataframe(
+                value=df[DEFAULT_NAMES],
+                headers=DEFAULT_NAMES,
+                type="pandas",
+                datatype=DATA_TITLE_TYPE,
+                interactive=False,
+                visible=True,
+                max_height=2400, 
             )
-
-    with gr.Row():
-        with gr.Accordion("📙 Citation", open=False):
-            citation_button = gr.Textbox(
-                value=CITATION_BUTTON_TEXT,
-                label=CITATION_BUTTON_LABEL,
-                lines=20,
-                elem_id="citation-button",
-                show_copy_button=True,
+            
+            refresh_button = gr.Button("Refresh")
+            
+            def update_with_tasks(*args):
+                return update_table(*args)
+
+            search_bar.change(
+                fn=update_with_tasks, 
+                inputs=[search_bar, min_size_slider, max_size_slider, tasks_select, type_select], 
+                outputs=data_component
             )
+            min_size_slider.change(
+                fn=update_with_tasks, 
+                inputs=[search_bar, min_size_slider, max_size_slider, tasks_select, type_select], 
+                outputs=data_component
+            )
+            max_size_slider.change(
+                fn=update_with_tasks, 
+                inputs=[search_bar, min_size_slider, max_size_slider, tasks_select, type_select], 
+                outputs=data_component
+            )
+            tasks_select.change(
+                fn=update_with_tasks, 
+                inputs=[search_bar, min_size_slider, max_size_slider, tasks_select, type_select], 
+                outputs=data_component
+            )
+            type_select.change(
+                fn=update_with_tasks, 
+                inputs=[search_bar, min_size_slider, max_size_slider, tasks_select, type_select], 
+                outputs=data_component
+            )
+            refresh_button.click(fn=refresh_data, outputs=data_component)
+            gr.Markdown(TABLE_INTRODUCTION)
+
+        # table 2
+        with gr.TabItem("📝 About", elem_id="qa-tab-table2", id=2):
+            gr.Markdown(LEADERBOARD_INFO, elem_classes="markdown-text")
+            gr.Image("dataset_statistics.png", width=900, label="Dataset Statistics")
+
+        # table 3
+        with gr.TabItem("🚀 Submit here! ", elem_id="submit-tab", id=3):
+            with gr.Row():
+                gr.Markdown(SUBMIT_INTRODUCTION, elem_classes="markdown-text")
 
-scheduler = BackgroundScheduler()
-scheduler.add_job(restart_space, "interval", seconds=1800)
-scheduler.start()
-demo.queue(default_concurrency_limit=40).launch()
+block.launch()

results.jsonl

@@ -0,0 +1,21 @@
+{"Models": "GPT-4o", "Model Size(B)": "-", "Frames": 256, "Type": "Proprietary", "URL": "https://openai.com/index/hello-gpt-4o/", "LP_Open": 39.4, "LP_MCQ": 64.8, "LR_Open": 23.1, "LR_MCQ": 62.6, "HP_Open": 26.4, "HP_MCQ": 42.1, "HR_Open": 29.2, "HR_MCQ": 50.4, "Overall_Open": 34.2, "Overall_MCQ": 59.5}
+{"Models": "Gemini-1.5-Flash", "Model Size(B)": "-", "Frames": 512, "Type": "Proprietary", "URL": "https://storage.googleapis.com/deepmind-media/gemini/gemini_v1_5_report.pdf", "LP_Open": 41.5, "LP_MCQ": 65.5, "LR_Open": 25.9, "LR_MCQ": 63.9, "HP_Open": 27.3, "HP_MCQ": 36.4, "HR_Open": 25.8, "HR_MCQ": 55.7, "Overall_Open": 35.1, "Overall_MCQ": 60.6}
+{"Models": "Gemini-2.5-Flash", "Model Size(B)": "-", "Frames": 256, "Type": "Proprietary", "URL": "https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/", "LP_Open": 42.4, "LP_MCQ": 64.1, "LR_Open": 30.6, "LR_MCQ": 65.3, "HP_Open": 25.6, "HP_MCQ": 33.9, "HR_Open": 26.9, "HR_MCQ": 54.2, "Overall_Open": 36.3, "Overall_MCQ": 59.3}
+{"Models": "Gemini-1.5-Pro", "Model Size(B)": "-", "Frames": 512, "Type": "Proprietary", "URL": "https://storage.googleapis.com/deepmind-media/gemini/gemini_v1_5_report.pdf", "LP_Open": 43.7, "LP_MCQ": 66.7, "LR_Open": 32.7, "LR_MCQ": 69.4, "HP_Open": 35.5, "HP_MCQ": 40.5, "HR_Open": 31.8, "HR_MCQ": 61.0, "Overall_Open": 39.3, "Overall_MCQ": 63.4}
+{"Models": "GPT-4.1-mini", "Model Size(B)": "-", "Frames": 256, "Type": "Proprietary", "URL": "https://openai.com/index/gpt-4-1/", "LP_Open": 46.0, "LP_MCQ": 68.6, "LR_Open": 32.0, "LR_MCQ": 68.7, "HP_Open": 27.3, "HP_MCQ": 38.8, "HR_Open": 32.6, "HR_MCQ": 57.6, "Overall_Open": 39.9, "Overall_MCQ": 63.5}
+{"Models": "GPT-4.1", "Model Size(B)": "-", "Frames": 256, "Type": "Proprietary", "URL": "https://openai.com/index/gpt-4-1/", "LP_Open": 47.2, "LP_MCQ": 68.8, "LR_Open": 29.9, "LR_MCQ": 68.7, "HP_Open": 28.1, "HP_MCQ": 38.0, "HR_Open": 34.5, "HR_MCQ": 59.5, "Overall_Open": 40.8, "Overall_MCQ": 64.0}
+{"Models": "Video-LLaVA", "Model Size(B)": "8", "Frames": 8, "Type": "Open-source", "URL": "https://github.com/PKU-YuanGroup/Video-LLaVA", "LP_Open": 13.2, "LP_MCQ": 27.5, "LR_Open": 6.1, "LR_MCQ": 33.3, "HP_Open": 14.0, "HP_MCQ": 24.8, "HR_Open": 6.1, "HR_MCQ": 26.5, "Overall_Open": 11.0, "Overall_MCQ": 27.7}
+{"Models": "Mantis-Idefics2", "Model Size(B)": "8", "Frames": 24, "Type": "Open-source", "URL": "https://arxiv.org/abs/2405.01483", "LP_Open": 17.8, "LP_MCQ": 33.2, "LR_Open": 9.5, "LR_MCQ": 29.9, "HP_Open": 16.5, "HP_MCQ": 16.5, "HR_Open": 8.3, "HR_MCQ": 29.9, "Overall_Open": 14.8, "Overall_MCQ": 30.6}
+{"Models": "LongVA", "Model Size(B)": "7", "Frames": 64, "Type": "Open-source", "URL": "https://arxiv.org/abs/2406.16852", "LP_Open": 20.5, "LP_MCQ": 43.3, "LR_Open": 6.8, "LR_MCQ": 33.3, "HP_Open": 19.0, "HP_MCQ": 24.0, "HR_Open": 9.5, "HR_MCQ": 31.8, "Overall_Open": 16.5, "Overall_MCQ": 38.0}
+{"Models": "Phi-4-Mini", "Model Size(B)": "5.6", "Frames": 128, "Type": "Open-source", "URL": "https://arxiv.org/abs/2503.01743", "LP_Open": 19.2, "LP_MCQ": 46.4, "LR_Open": 12.9, "LR_MCQ": 47.6, "HP_Open": 18.2, "HP_MCQ": 30.6, "HR_Open": 10.2, "HR_MCQ": 31.4, "Overall_Open": 16.5, "Overall_MCQ": 42.0}
+{"Models": "LongLLaVA", "Model Size(B)": "9", "Frames": 512, "Type": "Open-source", "URL": "https://huggingface.co/aws-prototyping/long-llava-qwen2-7b", "LP_Open": 21.7, "LP_MCQ": 41.2, "LR_Open": 15.0, "LR_MCQ": 34.0, "HP_Open": 14.0, "HP_MCQ": 29.8, "HR_Open": 10.2, "HR_MCQ": 29.2, "Overall_Open": 17.8, "Overall_MCQ": 36.9}
+{"Models": "Video-XL", "Model Size(B)": "7", "Frames": 512, "Type": "Open-source", "URL": "https://github.com/VectorSpaceLab/Video-XL", "LP_Open": 22.3, "LP_MCQ": 41.9, "LR_Open": 15.0, "LR_MCQ": 34.0, "HP_Open": 18.2, "HP_MCQ": 28.1, "HR_Open": 10.2, "HR_MCQ": 29.2, "Overall_Open": 18.6, "Overall_MCQ": 38.2}
+{"Models": "LongVU", "Model Size(B)": "7", "Frames": 512, "Type": "Open-source", "URL": "https://arxiv.org/abs/2410.17434", "LP_Open": 25.9, "LP_MCQ": 45.6, "LR_Open": 12.9, "LR_MCQ": 38.8, "HP_Open": 19.8, "HP_MCQ": 24.0, "HR_Open": 17.4, "HR_MCQ": 37.1, "Overall_Open": 22.1, "Overall_MCQ": 41.0}
+{"Models": "Vamba", "Model Size(B)": "10", "Frames": 512, "Type": "Open-source", "URL": "https://arxiv.org/abs/2503.11579", "LP_Open": 28.1, "LP_MCQ": 52.4, "LR_Open": 10.9, "LR_MCQ": 40.8, "HP_Open": 21.5, "HP_MCQ": 26.4, "HR_Open": 12.5, "HR_MCQ": 37.9, "Overall_Open": 22.3, "Overall_MCQ": 45.7}
+{"Models": "LLaVA-Video", "Model Size(B)": "7", "Frames": 64, "Type": "Open-source", "URL": "https://huggingface.co/lmms-lab/LLaVA-NeXT-Video-72B-Qwen2", "LP_Open": 28.5, "LP_MCQ": 53.5, "LR_Open": 13.6, "LR_MCQ": 47.6, "HP_Open": 20.7, "HP_MCQ": 28.9, "HR_Open": 19.3, "HR_MCQ": 40.2, "Overall_Open": 24.2, "Overall_MCQ": 47.8}
+{"Models": "InternVL2.5", "Model Size(B)": "8", "Frames": 64, "Type": "Open-source", "URL": "https://internvl.github.io/blog/2024-12-05-InternVL-2.5/", "LP_Open": 28.8, "LP_MCQ": 54.3, "LR_Open": 19.7, "LR_MCQ": 46.3, "HP_Open": 21.5, "HP_MCQ": 35.5, "HR_Open": 16.7, "HR_MCQ": 39.0, "Overall_Open": 24.6, "Overall_MCQ": 48.5}
+{"Models": "InternVL3", "Model Size(B)": "8", "Frames": 64, "Type": "Open-source", "URL": "https://arxiv.org/abs/2504.10479", "LP_Open": 30.3, "LP_MCQ": 54.6, "LR_Open": 17.0, "LR_MCQ": 49.0, "HP_Open": 24.0, "HP_MCQ": 34.7, "HR_Open": 13.3, "HR_MCQ": 36.7, "Overall_Open": 24.7, "Overall_MCQ": 48.4}
+{"Models": "Qwen2-VL", "Model Size(B)": "7", "Frames": 512, "Type": "Open-source", "URL": "https://github.com/QwenLM/Qwen2-VL", "LP_Open": 31.7, "LP_MCQ": 59.3, "LR_Open": 14.3, "LR_MCQ": 51.7, "HP_Open": 21.5, "HP_MCQ": 28.1, "HR_Open": 20.5, "HR_MCQ": 39.0, "Overall_Open": 26.5, "Overall_MCQ": 48.2}
+{"Models": "InternVideo2.5", "Model Size(B)": "8", "Frames": 512, "Type": "Open-source", "URL": "https://arxiv.org/abs/2501.12386", "LP_Open": 33.6, "LP_MCQ": 59.8, "LR_Open": 17.0, "LR_MCQ": 47.6, "HP_Open": 19.8, "HP_MCQ": 34.7, "HR_Open": 18.2, "HR_MCQ": 45.8, "Overall_Open": 27.2, "Overall_MCQ": 53.2}
+{"Models": "VideoChat-Flash", "Model Size(B)": "7", "Frames": 512, "Type": "Open-source", "URL": "https://github.com/OpenGVLab/VideoChat-Flash", "LP_Open": 33.3, "LP_MCQ": 57.7, "LR_Open": 16.3, "LR_MCQ": 43.5, "HP_Open": 21.5, "HP_MCQ": 33.9, "HR_Open": 17.4, "HR_MCQ": 44.7, "Overall_Open": 27.0, "Overall_MCQ": 51.2}
+{"Models": "Qwen2.5-VL", "Model Size(B)": "7", "Frames": 512, "Type": "Open-source", "URL": "https://arxiv.org/abs/2502.13923", "LP_Open": 33.9, "LP_MCQ": 51.7, "LR_Open": 15.6, "LR_MCQ": 48.3, "HP_Open": 24.8, "HP_MCQ": 31.4, "HR_Open": 17.8, "HR_MCQ": 39.8, "Overall_Open": 27.7, "Overall_MCQ": 46.9}

src/about.py

@@ -1,72 +0,0 @@
-from dataclasses import dataclass
-from enum import Enum
-
-@dataclass
-class Task:
-    benchmark: str
-    metric: str
-    col_name: str
-
-
-# Select your tasks here
-# ---------------------------------------------------
-class Tasks(Enum):
-    # task_key in the json file, metric_key in the json file, name to display in the leaderboard 
-    task0 = Task("anli_r1", "acc", "ANLI")
-    task1 = Task("logiqa", "acc_norm", "LogiQA")
-
-NUM_FEWSHOT = 0 # Change with your few shot
-# ---------------------------------------------------
-
-
-
-# Your leaderboard name
-TITLE = """<h1 align="center" id="space-title">Demo leaderboard</h1>"""
-
-# What does your leaderboard evaluate?
-INTRODUCTION_TEXT = """
-Intro text
-"""
-
-# Which evaluations are you running? how can people reproduce what you have?
-LLM_BENCHMARKS_TEXT = f"""
-## How it works
-
-## Reproducibility
-To reproduce our results, here is the commands you can run:
-
-"""
-
-EVALUATION_QUEUE_TEXT = """
-## Some good practices before submitting a model
-
-### 1) Make sure you can load your model and tokenizer using AutoClasses:
-```python
-from transformers import AutoConfig, AutoModel, AutoTokenizer
-config = AutoConfig.from_pretrained("your model name", revision=revision)
-model = AutoModel.from_pretrained("your model name", revision=revision)
-tokenizer = AutoTokenizer.from_pretrained("your model name", revision=revision)
-```
-If this step fails, follow the error messages to debug your model before submitting it. It's likely your model has been improperly uploaded.
-
-Note: make sure your model is public!
-Note: if your model needs `use_remote_code=True`, we do not support this option yet but we are working on adding it, stay posted!
-
-### 2) Convert your model weights to [safetensors](https://huggingface.co/docs/safetensors/index)
-It's a new format for storing weights which is safer and faster to load and use. It will also allow us to add the number of parameters of your model to the `Extended Viewer`!
-
-### 3) Make sure your model has an open license!
-This is a leaderboard for Open LLMs, and we'd love for as many people as possible to know they can use your model 🤗
-
-### 4) Fill up your model card
-When we add extra information about models to the leaderboard, it will be automatically taken from the model card
-
-## In case of model failure
-If your model is displayed in the `FAILED` category, its execution stopped.
-Make sure you have followed the above steps first.
-If everything is done, check you can launch the EleutherAIHarness on your model locally, using the above command without modifications (you can add `--limit` to limit the number of examples per task).
-"""
-
-CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
-CITATION_BUTTON_TEXT = r"""
-"""

src/display/css_html_js.py

@@ -1,105 +0,0 @@
-custom_css = """
-
-.markdown-text {
-    font-size: 16px !important;
-}
-
-#models-to-add-text {
-    font-size: 18px !important;
-}
-
-#citation-button span {
-    font-size: 16px !important;
-}
-
-#citation-button textarea {
-    font-size: 16px !important;
-}
-
-#citation-button > label > button {
-    margin: 6px;
-    transform: scale(1.3);
-}
-
-#leaderboard-table {
-    margin-top: 15px
-}
-
-#leaderboard-table-lite {
-    margin-top: 15px
-}
-
-#search-bar-table-box > div:first-child {
-    background: none;
-    border: none;
-}
- 
-#search-bar {
-    padding: 0px;
-}
-
-/* Limit the width of the first AutoEvalColumn so that names don't expand too much */
-#leaderboard-table td:nth-child(2),
-#leaderboard-table th:nth-child(2) {
-    max-width: 400px;
-    overflow: auto;
-    white-space: nowrap;
-}
-
-.tab-buttons button {
-    font-size: 20px;
-}
-
-#scale-logo {
-    border-style: none !important;
-    box-shadow: none;
-    display: block;
-    margin-left: auto;
-    margin-right: auto;
-    max-width: 600px;
-}
-
-#scale-logo .download {
-    display: none;
-}
-#filter_type{
-    border: 0;
-    padding-left: 0;
-    padding-top: 0;
-}
-#filter_type label {
-    display: flex;
-}
-#filter_type label > span{
-    margin-top: var(--spacing-lg);
-    margin-right: 0.5em;
-}
-#filter_type label > .wrap{
-    width: 103px;
-}
-#filter_type label > .wrap .wrap-inner{  
-    padding: 2px;
-}
-#filter_type label > .wrap .wrap-inner input{
-    width: 1px
-}
-#filter-columns-type{
-    border:0;
-    padding:0.5;
-}
-#filter-columns-size{
-    border:0;
-    padding:0.5;
-}
-#box-filter > .form{
-    border: 0
-}
-"""
-
-get_window_url_params = """
-    function(url_params) {
-        const params = new URLSearchParams(window.location.search);
-        url_params = Object.fromEntries(params);
-        return url_params;
-    }
-    """

src/display/formatting.py

@@ -1,27 +0,0 @@
-def model_hyperlink(link, model_name):
-    return f'<a target="_blank" href="{link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;">{model_name}</a>'
-
-
-def make_clickable_model(model_name):
-    link = f"https://huggingface.co/{model_name}"
-    return model_hyperlink(link, model_name)
-
-
-def styled_error(error):
-    return f"<p style='color: red; font-size: 20px; text-align: center;'>{error}</p>"
-
-
-def styled_warning(warn):
-    return f"<p style='color: orange; font-size: 20px; text-align: center;'>{warn}</p>"
-
-
-def styled_message(message):
-    return f"<p style='color: green; font-size: 20px; text-align: center;'>{message}</p>"
-
-
-def has_no_nan_values(df, columns):
-    return df[columns].notna().all(axis=1)
-
-
-def has_nan_values(df, columns):
-    return df[columns].isna().any(axis=1)

src/display/utils.py

@@ -1,110 +0,0 @@
-from dataclasses import dataclass, make_dataclass
-from enum import Enum
-
-import pandas as pd
-
-from src.about import Tasks
-
-def fields(raw_class):
-    return [v for k, v in raw_class.__dict__.items() if k[:2] != "__" and k[-2:] != "__"]
-
-
-# These classes are for user facing column names,
-# to avoid having to change them all around the code
-# when a modif is needed
-@dataclass
-class ColumnContent:
-    name: str
-    type: str
-    displayed_by_default: bool
-    hidden: bool = False
-    never_hidden: bool = False
-
-## Leaderboard columns
-auto_eval_column_dict = []
-# Init
-auto_eval_column_dict.append(["model_type_symbol", ColumnContent, ColumnContent("T", "str", True, never_hidden=True)])
-auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])
-#Scores
-auto_eval_column_dict.append(["average", ColumnContent, ColumnContent("Average ⬆️", "number", True)])
-for task in Tasks:
-    auto_eval_column_dict.append([task.name, ColumnContent, ColumnContent(task.value.col_name, "number", True)])
-# Model information
-auto_eval_column_dict.append(["model_type", ColumnContent, ColumnContent("Type", "str", False)])
-auto_eval_column_dict.append(["architecture", ColumnContent, ColumnContent("Architecture", "str", False)])
-auto_eval_column_dict.append(["weight_type", ColumnContent, ColumnContent("Weight type", "str", False, True)])
-auto_eval_column_dict.append(["precision", ColumnContent, ColumnContent("Precision", "str", False)])
-auto_eval_column_dict.append(["license", ColumnContent, ColumnContent("Hub License", "str", False)])
-auto_eval_column_dict.append(["params", ColumnContent, ColumnContent("#Params (B)", "number", False)])
-auto_eval_column_dict.append(["likes", ColumnContent, ColumnContent("Hub ❤️", "number", False)])
-auto_eval_column_dict.append(["still_on_hub", ColumnContent, ColumnContent("Available on the hub", "bool", False)])
-auto_eval_column_dict.append(["revision", ColumnContent, ColumnContent("Model sha", "str", False, False)])
-
-# We use make dataclass to dynamically fill the scores from Tasks
-AutoEvalColumn = make_dataclass("AutoEvalColumn", auto_eval_column_dict, frozen=True)
-
-## For the queue columns in the submission tab
-@dataclass(frozen=True)
-class EvalQueueColumn:  # Queue column
-    model = ColumnContent("model", "markdown", True)
-    revision = ColumnContent("revision", "str", True)
-    private = ColumnContent("private", "bool", True)
-    precision = ColumnContent("precision", "str", True)
-    weight_type = ColumnContent("weight_type", "str", "Original")
-    status = ColumnContent("status", "str", True)
-
-## All the model information that we might need
-@dataclass
-class ModelDetails:
-    name: str
-    display_name: str = ""
-    symbol: str = "" # emoji
-
-
-class ModelType(Enum):
-    PT = ModelDetails(name="pretrained", symbol="🟢")
-    FT = ModelDetails(name="fine-tuned", symbol="🔶")
-    IFT = ModelDetails(name="instruction-tuned", symbol="⭕")
-    RL = ModelDetails(name="RL-tuned", symbol="🟦")
-    Unknown = ModelDetails(name="", symbol="?")
-
-    def to_str(self, separator=" "):
-        return f"{self.value.symbol}{separator}{self.value.name}"
-
-    @staticmethod
-    def from_str(type):
-        if "fine-tuned" in type or "🔶" in type:
-            return ModelType.FT
-        if "pretrained" in type or "🟢" in type:
-            return ModelType.PT
-        if "RL-tuned" in type or "🟦" in type:
-            return ModelType.RL
-        if "instruction-tuned" in type or "⭕" in type:
-            return ModelType.IFT
-        return ModelType.Unknown
-
-class WeightType(Enum):
-    Adapter = ModelDetails("Adapter")
-    Original = ModelDetails("Original")
-    Delta = ModelDetails("Delta")
-
-class Precision(Enum):
-    float16 = ModelDetails("float16")
-    bfloat16 = ModelDetails("bfloat16")
-    Unknown = ModelDetails("?")
-
-    def from_str(precision):
-        if precision in ["torch.float16", "float16"]:
-            return Precision.float16
-        if precision in ["torch.bfloat16", "bfloat16"]:
-            return Precision.bfloat16
-        return Precision.Unknown
-
-# Column selection
-COLS = [c.name for c in fields(AutoEvalColumn) if not c.hidden]
-
-EVAL_COLS = [c.name for c in fields(EvalQueueColumn)]
-EVAL_TYPES = [c.type for c in fields(EvalQueueColumn)]
-
-BENCHMARK_COLS = [t.value.col_name for t in Tasks]
-

src/envs.py

@@ -1,25 +0,0 @@
-import os
-
-from huggingface_hub import HfApi
-
-# Info to change for your repository
-# ----------------------------------
-TOKEN = os.environ.get("HF_TOKEN") # A read/write token for your org
-
-OWNER = "demo-leaderboard-backend" # Change to your org - don't forget to create a results and request dataset, with the correct format!
-# ----------------------------------
-
-REPO_ID = f"{OWNER}/leaderboard"
-QUEUE_REPO = f"{OWNER}/requests"
-RESULTS_REPO = f"{OWNER}/results"
-
-# If you setup a cache later, just change HF_HOME
-CACHE_PATH=os.getenv("HF_HOME", ".")
-
-# Local caches
-EVAL_REQUESTS_PATH = os.path.join(CACHE_PATH, "eval-queue")
-EVAL_RESULTS_PATH = os.path.join(CACHE_PATH, "eval-results")
-EVAL_REQUESTS_PATH_BACKEND = os.path.join(CACHE_PATH, "eval-queue-bk")
-EVAL_RESULTS_PATH_BACKEND = os.path.join(CACHE_PATH, "eval-results-bk")
-
-API = HfApi(token=TOKEN)

src/leaderboard/read_evals.py

@@ -1,196 +0,0 @@
-import glob
-import json
-import math
-import os
-from dataclasses import dataclass
-
-import dateutil
-import numpy as np
-
-from src.display.formatting import make_clickable_model
-from src.display.utils import AutoEvalColumn, ModelType, Tasks, Precision, WeightType
-from src.submission.check_validity import is_model_on_hub
-
-
-@dataclass
-class EvalResult:
-    """Represents one full evaluation. Built from a combination of the result and request file for a given run.
-    """
-    eval_name: str # org_model_precision (uid)
-    full_model: str # org/model (path on hub)
-    org: str 
-    model: str
-    revision: str # commit hash, "" if main
-    results: dict
-    precision: Precision = Precision.Unknown
-    model_type: ModelType = ModelType.Unknown # Pretrained, fine tuned, ...
-    weight_type: WeightType = WeightType.Original # Original or Adapter
-    architecture: str = "Unknown" 
-    license: str = "?"
-    likes: int = 0
-    num_params: int = 0
-    date: str = "" # submission date of request file
-    still_on_hub: bool = False
-
-    @classmethod
-    def init_from_json_file(self, json_filepath):
-        """Inits the result from the specific model result file"""
-        with open(json_filepath) as fp:
-            data = json.load(fp)
-
-        config = data.get("config")
-
-        # Precision
-        precision = Precision.from_str(config.get("model_dtype"))
-
-        # Get model and org
-        org_and_model = config.get("model_name", config.get("model_args", None))
-        org_and_model = org_and_model.split("/", 1)
-
-        if len(org_and_model) == 1:
-            org = None
-            model = org_and_model[0]
-            result_key = f"{model}_{precision.value.name}"
-        else:
-            org = org_and_model[0]
-            model = org_and_model[1]
-            result_key = f"{org}_{model}_{precision.value.name}"
-        full_model = "/".join(org_and_model)
-
-        still_on_hub, _, model_config = is_model_on_hub(
-            full_model, config.get("model_sha", "main"), trust_remote_code=True, test_tokenizer=False
-        )
-        architecture = "?"
-        if model_config is not None:
-            architectures = getattr(model_config, "architectures", None)
-            if architectures:
-                architecture = ";".join(architectures)
-
-        # Extract results available in this file (some results are split in several files)
-        results = {}
-        for task in Tasks:
-            task = task.value
-
-            # We average all scores of a given metric (not all metrics are present in all files)
-            accs = np.array([v.get(task.metric, None) for k, v in data["results"].items() if task.benchmark == k])
-            if accs.size == 0 or any([acc is None for acc in accs]):
-                continue
-
-            mean_acc = np.mean(accs) * 100.0
-            results[task.benchmark] = mean_acc
-
-        return self(
-            eval_name=result_key,
-            full_model=full_model,
-            org=org,
-            model=model,
-            results=results,
-            precision=precision,  
-            revision= config.get("model_sha", ""),
-            still_on_hub=still_on_hub,
-            architecture=architecture
-        )
-
-    def update_with_request_file(self, requests_path):
-        """Finds the relevant request file for the current model and updates info with it"""
-        request_file = get_request_file_for_model(requests_path, self.full_model, self.precision.value.name)
-
-        try:
-            with open(request_file, "r") as f:
-                request = json.load(f)
-            self.model_type = ModelType.from_str(request.get("model_type", ""))
-            self.weight_type = WeightType[request.get("weight_type", "Original")]
-            self.license = request.get("license", "?")
-            self.likes = request.get("likes", 0)
-            self.num_params = request.get("params", 0)
-            self.date = request.get("submitted_time", "")
-        except Exception:
-            print(f"Could not find request file for {self.org}/{self.model} with precision {self.precision.value.name}")
-
-    def to_dict(self):
-        """Converts the Eval Result to a dict compatible with our dataframe display"""
-        average = sum([v for v in self.results.values() if v is not None]) / len(Tasks)
-        data_dict = {
-            "eval_name": self.eval_name,  # not a column, just a save name,
-            AutoEvalColumn.precision.name: self.precision.value.name,
-            AutoEvalColumn.model_type.name: self.model_type.value.name,
-            AutoEvalColumn.model_type_symbol.name: self.model_type.value.symbol,
-            AutoEvalColumn.weight_type.name: self.weight_type.value.name,
-            AutoEvalColumn.architecture.name: self.architecture,
-            AutoEvalColumn.model.name: make_clickable_model(self.full_model),
-            AutoEvalColumn.revision.name: self.revision,
-            AutoEvalColumn.average.name: average,
-            AutoEvalColumn.license.name: self.license,
-            AutoEvalColumn.likes.name: self.likes,
-            AutoEvalColumn.params.name: self.num_params,
-            AutoEvalColumn.still_on_hub.name: self.still_on_hub,
-        }
-
-        for task in Tasks:
-            data_dict[task.value.col_name] = self.results[task.value.benchmark]
-
-        return data_dict
-
-
-def get_request_file_for_model(requests_path, model_name, precision):
-    """Selects the correct request file for a given model. Only keeps runs tagged as FINISHED"""
-    request_files = os.path.join(
-        requests_path,
-        f"{model_name}_eval_request_*.json",
-    )
-    request_files = glob.glob(request_files)
-
-    # Select correct request file (precision)
-    request_file = ""
-    request_files = sorted(request_files, reverse=True)
-    for tmp_request_file in request_files:
-        with open(tmp_request_file, "r") as f:
-            req_content = json.load(f)
-            if (
-                req_content["status"] in ["FINISHED"]
-                and req_content["precision"] == precision.split(".")[-1]
-            ):
-                request_file = tmp_request_file
-    return request_file
-
-
-def get_raw_eval_results(results_path: str, requests_path: str) -> list[EvalResult]:
-    """From the path of the results folder root, extract all needed info for results"""
-    model_result_filepaths = []
-
-    for root, _, files in os.walk(results_path):
-        # We should only have json files in model results
-        if len(files) == 0 or any([not f.endswith(".json") for f in files]):
-            continue
-
-        # Sort the files by date
-        try:
-            files.sort(key=lambda x: x.removesuffix(".json").removeprefix("results_")[:-7])
-        except dateutil.parser._parser.ParserError:
-            files = [files[-1]]
-
-        for file in files:
-            model_result_filepaths.append(os.path.join(root, file))
-
-    eval_results = {}
-    for model_result_filepath in model_result_filepaths:
-        # Creation of result
-        eval_result = EvalResult.init_from_json_file(model_result_filepath)
-        eval_result.update_with_request_file(requests_path)
-
-        # Store results of same eval together
-        eval_name = eval_result.eval_name
-        if eval_name in eval_results.keys():
-            eval_results[eval_name].results.update({k: v for k, v in eval_result.results.items() if v is not None})
-        else:
-            eval_results[eval_name] = eval_result
-
-    results = []
-    for v in eval_results.values():
-        try:
-            v.to_dict() # we test if the dict version is complete
-            results.append(v)
-        except KeyError:  # not all eval values present
-            continue
-
-    return results

src/populate.py

@@ -1,58 +0,0 @@
-import json
-import os
-
-import pandas as pd
-
-from src.display.formatting import has_no_nan_values, make_clickable_model
-from src.display.utils import AutoEvalColumn, EvalQueueColumn
-from src.leaderboard.read_evals import get_raw_eval_results
-
-
-def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame:
-    """Creates a dataframe from all the individual experiment results"""
-    raw_data = get_raw_eval_results(results_path, requests_path)
-    all_data_json = [v.to_dict() for v in raw_data]
-
-    df = pd.DataFrame.from_records(all_data_json)
-    df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
-    df = df[cols].round(decimals=2)
-
-    # filter out if any of the benchmarks have not been produced
-    df = df[has_no_nan_values(df, benchmark_cols)]
-    return df
-
-
-def get_evaluation_queue_df(save_path: str, cols: list) -> list[pd.DataFrame]:
-    """Creates the different dataframes for the evaluation queues requestes"""
-    entries = [entry for entry in os.listdir(save_path) if not entry.startswith(".")]
-    all_evals = []
-
-    for entry in entries:
-        if ".json" in entry:
-            file_path = os.path.join(save_path, entry)
-            with open(file_path) as fp:
-                data = json.load(fp)
-
-            data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])
-            data[EvalQueueColumn.revision.name] = data.get("revision", "main")
-
-            all_evals.append(data)
-        elif ".md" not in entry:
-            # this is a folder
-            sub_entries = [e for e in os.listdir(f"{save_path}/{entry}") if os.path.isfile(e) and not e.startswith(".")]
-            for sub_entry in sub_entries:
-                file_path = os.path.join(save_path, entry, sub_entry)
-                with open(file_path) as fp:
-                    data = json.load(fp)
-
-                data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])
-                data[EvalQueueColumn.revision.name] = data.get("revision", "main")
-                all_evals.append(data)
-
-    pending_list = [e for e in all_evals if e["status"] in ["PENDING", "RERUN"]]
-    running_list = [e for e in all_evals if e["status"] == "RUNNING"]
-    finished_list = [e for e in all_evals if e["status"].startswith("FINISHED") or e["status"] == "PENDING_NEW_EVAL"]
-    df_pending = pd.DataFrame.from_records(pending_list, columns=cols)
-    df_running = pd.DataFrame.from_records(running_list, columns=cols)
-    df_finished = pd.DataFrame.from_records(finished_list, columns=cols)
-    return df_finished[cols], df_running[cols], df_pending[cols]

src/submission/check_validity.py

@@ -1,99 +0,0 @@
-import json
-import os
-import re
-from collections import defaultdict
-from datetime import datetime, timedelta, timezone
-
-import huggingface_hub
-from huggingface_hub import ModelCard
-from huggingface_hub.hf_api import ModelInfo
-from transformers import AutoConfig
-from transformers.models.auto.tokenization_auto import AutoTokenizer
-
-def check_model_card(repo_id: str) -> tuple[bool, str]:
-    """Checks if the model card and license exist and have been filled"""
-    try:
-        card = ModelCard.load(repo_id)
-    except huggingface_hub.utils.EntryNotFoundError:
-        return False, "Please add a model card to your model to explain how you trained/fine-tuned it."
-
-    # Enforce license metadata
-    if card.data.license is None:
-        if not ("license_name" in card.data and "license_link" in card.data):
-            return False, (
-                "License not found. Please add a license to your model card using the `license` metadata or a"
-                " `license_name`/`license_link` pair."
-            )
-
-    # Enforce card content
-    if len(card.text) < 200:
-        return False, "Please add a description to your model card, it is too short."
-
-    return True, ""
-
-def is_model_on_hub(model_name: str, revision: str, token: str = None, trust_remote_code=False, test_tokenizer=False) -> tuple[bool, str]:
-    """Checks if the model model_name is on the hub, and whether it (and its tokenizer) can be loaded with AutoClasses."""
-    try:
-        config = AutoConfig.from_pretrained(model_name, revision=revision, trust_remote_code=trust_remote_code, token=token)
-        if test_tokenizer:
-            try:
-                tk = AutoTokenizer.from_pretrained(model_name, revision=revision, trust_remote_code=trust_remote_code, token=token)
-            except ValueError as e:
-                return (
-                    False,
-                    f"uses a tokenizer which is not in a transformers release: {e}",
-                    None
-                )
-            except Exception as e:
-                return (False, "'s tokenizer cannot be loaded. Is your tokenizer class in a stable transformers release, and correctly configured?", None)
-        return True, None, config
-
-    except ValueError:
-        return (
-            False,
-            "needs to be launched with `trust_remote_code=True`. For safety reason, we do not allow these models to be automatically submitted to the leaderboard.",
-            None
-        )
-
-    except Exception as e:
-        return False, "was not found on hub!", None
-
-
-def get_model_size(model_info: ModelInfo, precision: str):
-    """Gets the model size from the configuration, or the model name if the configuration does not contain the information."""
-    try:
-        model_size = round(model_info.safetensors["total"] / 1e9, 3)
-    except (AttributeError, TypeError):
-        return 0  # Unknown model sizes are indicated as 0, see NUMERIC_INTERVALS in app.py
-
-    size_factor = 8 if (precision == "GPTQ" or "gptq" in model_info.modelId.lower()) else 1
-    model_size = size_factor * model_size
-    return model_size
-
-def get_model_arch(model_info: ModelInfo):
-    """Gets the model architecture from the configuration"""
-    return model_info.config.get("architectures", "Unknown")
-
-def already_submitted_models(requested_models_dir: str) -> set[str]:
-    """Gather a list of already submitted models to avoid duplicates"""
-    depth = 1
-    file_names = []
-    users_to_submission_dates = defaultdict(list)
-
-    for root, _, files in os.walk(requested_models_dir):
-        current_depth = root.count(os.sep) - requested_models_dir.count(os.sep)
-        if current_depth == depth:
-            for file in files:
-                if not file.endswith(".json"):
-                    continue
-                with open(os.path.join(root, file), "r") as f:
-                    info = json.load(f)
-                    file_names.append(f"{info['model']}_{info['revision']}_{info['precision']}")
-
-                    # Select organisation
-                    if info["model"].count("/") == 0 or "submitted_time" not in info:
-                        continue
-                    organisation, _ = info["model"].split("/")
-                    users_to_submission_dates[organisation].append(info["submitted_time"])
-
-    return set(file_names), users_to_submission_dates

src/submission/submit.py

@@ -1,119 +0,0 @@
-import json
-import os
-from datetime import datetime, timezone
-
-from src.display.formatting import styled_error, styled_message, styled_warning
-from src.envs import API, EVAL_REQUESTS_PATH, TOKEN, QUEUE_REPO
-from src.submission.check_validity import (
-    already_submitted_models,
-    check_model_card,
-    get_model_size,
-    is_model_on_hub,
-)
-
-REQUESTED_MODELS = None
-USERS_TO_SUBMISSION_DATES = None
-
-def add_new_eval(
-    model: str,
-    base_model: str,
-    revision: str,
-    precision: str,
-    weight_type: str,
-    model_type: str,
-):
-    global REQUESTED_MODELS
-    global USERS_TO_SUBMISSION_DATES
-    if not REQUESTED_MODELS:
-        REQUESTED_MODELS, USERS_TO_SUBMISSION_DATES = already_submitted_models(EVAL_REQUESTS_PATH)
-
-    user_name = ""
-    model_path = model
-    if "/" in model:
-        user_name = model.split("/")[0]
-        model_path = model.split("/")[1]
-
-    precision = precision.split(" ")[0]
-    current_time = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
-
-    if model_type is None or model_type == "":
-        return styled_error("Please select a model type.")
-
-    # Does the model actually exist?
-    if revision == "":
-        revision = "main"
-
-    # Is the model on the hub?
-    if weight_type in ["Delta", "Adapter"]:
-        base_model_on_hub, error, _ = is_model_on_hub(model_name=base_model, revision=revision, token=TOKEN, test_tokenizer=True)
-        if not base_model_on_hub:
-            return styled_error(f'Base model "{base_model}" {error}')
-
-    if not weight_type == "Adapter":
-        model_on_hub, error, _ = is_model_on_hub(model_name=model, revision=revision, token=TOKEN, test_tokenizer=True)
-        if not model_on_hub:
-            return styled_error(f'Model "{model}" {error}')
-
-    # Is the model info correctly filled?
-    try:
-        model_info = API.model_info(repo_id=model, revision=revision)
-    except Exception:
-        return styled_error("Could not get your model information. Please fill it up properly.")
-
-    model_size = get_model_size(model_info=model_info, precision=precision)
-
-    # Were the model card and license filled?
-    try:
-        license = model_info.cardData["license"]
-    except Exception:
-        return styled_error("Please select a license for your model")
-
-    modelcard_OK, error_msg = check_model_card(model)
-    if not modelcard_OK:
-        return styled_error(error_msg)
-
-    # Seems good, creating the eval
-    print("Adding new eval")
-
-    eval_entry = {
-        "model": model,
-        "base_model": base_model,
-        "revision": revision,
-        "precision": precision,
-        "weight_type": weight_type,
-        "status": "PENDING",
-        "submitted_time": current_time,
-        "model_type": model_type,
-        "likes": model_info.likes,
-        "params": model_size,
-        "license": license,
-        "private": False,
-    }
-
-    # Check for duplicate submission
-    if f"{model}_{revision}_{precision}" in REQUESTED_MODELS:
-        return styled_warning("This model has been already submitted.")
-
-    print("Creating eval file")
-    OUT_DIR = f"{EVAL_REQUESTS_PATH}/{user_name}"
-    os.makedirs(OUT_DIR, exist_ok=True)
-    out_path = f"{OUT_DIR}/{model_path}_eval_request_False_{precision}_{weight_type}.json"
-
-    with open(out_path, "w") as f:
-        f.write(json.dumps(eval_entry))
-
-    print("Uploading eval file")
-    API.upload_file(
-        path_or_fileobj=out_path,
-        path_in_repo=out_path.split("eval-queue/")[1],
-        repo_id=QUEUE_REPO,
-        repo_type="dataset",
-        commit_message=f"Add {model} to eval queue",
-    )
-
-    # Remove the local file
-    os.remove(out_path)
-
-    return styled_message(
-        "Your request has been submitted to the evaluation queue!\nPlease wait for up to an hour for the model to show in the PENDING list."
-    )

utils.py

@@ -0,0 +1,170 @@
+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)", "Type", "Frames"]
+
+TASKS = ["LP_Open", "LP_MCQ", "LR_Open", "LR_MCQ", "HP_Open", "HP_MCQ", "HR_Open", "HR_MCQ", "Overall_Open", "Overall_MCQ"]
+OPEN_TASKS = ["LP_Open", "LR_Open", "HP_Open", "HR_Open", "Overall_Open"]
+MCQ_TASKS = ["LP_MCQ", "LR_MCQ", "HP_MCQ", "HR_MCQ", "Overall_MCQ"]
+
+DEFAULT_NAMES = BASE_COLS + OPEN_TASKS
+COLUMN_NAMES = BASE_COLS + TASKS
+
+GROUP_FIELD = "Type"  # "Proprietary" or "Open-source"
+
+DATA_TITLE_TYPE = ['number', 'markdown', 'str', 'number', 'str', 'str'] + \
+                    ['number'] * len(TASKS)
+
+LEADERBOARD_INTRODUCTION = """
+# 🥇 **VideoEval-Pro Leaderboard**
+### A More Robust and Realistic QA Evaluation benchmark of Multi-modal LLMs in long video understanding
+## Introduction
+Do existing long video benchmarks faithfully reflect model's real capacity to understand long video content? Do the gains reported by newer models genuinely translate into stronger long video comprehension capability, or are they illusional? To probe these questions, we present VideoEval-Pro, a more robust and realistic long video understanding benchmark containing open-ended, short-answer QA problems. To construct VideoEval-Pro, we source the questions from four existing long video understanding MCQ benchmarks, and reformat these questions into free-form questions. We apply a series of filtering methods based on video duration, question and answer type, answerability and QA difficulty to ensure the quality of our benchmark. Our final benchmark contains a total of 1,289 short-answer questions based on 465 videos, with an average duration of 38 minutes. \n
+| [**📈Overview**](https://tiger-ai-lab.github.io/VideoEval-Pro) 
+| [**👨‍💻Github**](https://github.com/TIGER-AI-Lab/VideoEval-Pro) 
+| [**📖VideoEval-Pro Paper**](https://arxiv.org/abs/2505.14640) 
+| [**🤗HuggingFace**](https://huggingface.co/datasets/TIGER-Lab/VideoEval-Pro) |
+"""
+
+TABLE_INTRODUCTION = """Models are ranked based on Overall_Open."""
+
+LEADERBOARD_INFO = """
+## Dataset Statistics and Tasks Info
+* Local Perception (LP): LP focuses on identifying and retrieving visual elements or actions from a short video clip in a long video. Subtypes in this category include Segment QA, Needle-InA-Haystack (NIAH) QA, Attribute Perception, Action Recognition, Object Recognition, Entity Recognition, Key Information Retrieval and a combined Other subtype. 
+* Local Reasoning (LR): LR focuses on reasoning within short temporal windows, such as inferring causality, temporal order, or changes that happen over a local sequence of events. The four subtypes in this category are Egocentric Video Reasoning, Object Reasoning, Temporal Reasoning and Action Reasoning. 
+* Holistic Perception (HP): HP involves a global and holistic understanding of statistical, structural, or spatial information, typically requiring visual aggregation. In VIDEOEVAL-PRO, HP is comprised of Visual Counting problems. 
+* Holistic Reasoning (HR): HR requires abstract or high-level understanding of long videos across events or scenes, often involving narrative or intent understanding. The two subtypes for HR are Event Understanding and Plot Reasoning.
+"""
+
+CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
+CITATION_BUTTON_TEXT = r"""@misc{ma2025videoevalprorobustrealisticlong,
+      title={VideoEval-Pro: Robust and Realistic Long Video Understanding Evaluation}, 
+      author={Wentao Ma and Weiming Ren and Yiming Jia and Zhuofeng Li and Ping Nie and Ge Zhang and Wenhu Chen},
+      year={2025},
+      eprint={2505.14640},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2505.14640}, 
+}"""
+
+SUBMIT_INTRODUCTION = """# Submit on VideoEval-Pro Leaderboard Introduction
+## The evaluattion model should be used is *GPT-4o-0806*
+## ⚠ Please note that you need to submit the JSON file with the following format:
+```json
+[
+    {
+        "Models": "<Model Name>",
+        "Model Size(B)": "100 or -",
+        "Frames": "<Number of Frames>",
+        "Type": "Proprietary or Open-source",
+        "URL": "<Model URL>" or null,
+        "LP_Open": 50.0 or null,
+        "LP_MCQ": 50.0 or null,
+        "LR_Open": 50.0 or null,
+        "LR_MCQ": 50.0 or null,
+        "HP_Open": 50.0 or null,
+        "HP_MCQ": 50.0 or null,
+        "HR_Open": 50.0 or null,
+        "HR_MCQ": 50.0 or null,
+        "Overall_Open": 50.0,
+        "Overall_MCQ": 50.0,
+    }, 
+]
+```
+
+You may refer to the [**GitHub page**](https://github.com/TIGER-AI-Lab/VideoEval-Pro) for instructions about evaluating your model. \n
+Please send us an email at tonyyyma@gmail.com, 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)
+    for task in TASKS:
+        if df[task].isnull().any():
+            df[task] = df[task].apply(lambda score: '-' if pd.isna(score) else score)
+    df = df.sort_values(by=['Overall_Open'], ascending=False)
+    df = create_hyperlinked_names(df)
+    df['Rank'] = range(1, len(df) + 1)
+    return df
+
+def refresh_data():
+    df = get_df()
+    return df[DEFAULT_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 <= 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)', 'Frames', 'Type', 'Overall_Open']
+    selected_columns = base_columns + selected_tasks
+    
+    available_columns = [col for col in selected_columns if col in df.columns]
+    return df[available_columns]