init

.gitattributes

@@ -25,7 +25,6 @@
 *.safetensors filter=lfs diff=lfs merge=lfs -text
 saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
 *.tflite filter=lfs diff=lfs merge=lfs -text
 *.tgz filter=lfs diff=lfs merge=lfs -text
 *.wasm filter=lfs diff=lfs merge=lfs -text
@@ -33,3 +32,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip 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

.gitignore

@@ -0,0 +1,13 @@
+auto_evals/
+venv/
+__pycache__/
+.env
+.ipynb_checkpoints
+*ipynb
+.vscode/
+
+eval-queue/
+eval-results/
+eval-queue-bk/
+eval-results-bk/
+logs/

.pre-commit-config.yaml

@@ -0,0 +1,53 @@
+# Copyright (c) 2022, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+default_language_version:
+  python: python3
+
+ci:
+  autofix_prs: true
+  autoupdate_commit_msg: '[pre-commit.ci] pre-commit suggestions'
+  autoupdate_schedule: quarterly
+
+repos:
+  - repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v4.3.0
+    hooks:
+      - id: check-yaml
+      - id: check-case-conflict
+      - id: detect-private-key
+      - id: check-added-large-files
+        args: ['--maxkb=1000']
+      - id: requirements-txt-fixer
+      - id: end-of-file-fixer
+      - id: trailing-whitespace
+
+  - repo: https://github.com/PyCQA/isort
+    rev: 5.12.0
+    hooks:
+      - id: isort
+        name: Format imports
+
+  - repo: https://github.com/psf/black
+    rev: 22.12.0
+    hooks:
+      - id: black
+        name: Format code
+        additional_dependencies: ['click==8.0.2']
+
+  - repo: https://github.com/charliermarsh/ruff-pre-commit
+    # Ruff version.
+    rev: 'v0.0.267'
+    hooks:
+      - id: ruff

Makefile

@@ -0,0 +1,13 @@
+.PHONY: style format
+
+
+style:
+	python -m black --line-length 119 .
+	python -m isort .
+	ruff check --fix .
+
+
+quality:
+	python -m black --check --line-length 119 .
+	python -m isort --check-only .
+	ruff check .

README.md

@@ -1,13 +1,44 @@
 ---
-title: Unlearned DM Benchmark
-emoji: 🌖
+title: UnlearnDiffAtk Benchmark
+emoji: 🥇
 colorFrom: green
-colorTo: gray
+colorTo: indigo
 sdk: gradio
-sdk_version: 4.36.1
 app_file: app.py
-pinned: false
+pinned: true
 license: apache-2.0
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Start the configuration
+
+Most of the variables to change for a default leaderboard are in `src/env.py` (replace the path for your leaderboard) and `src/about.py` (for tasks).
+
+Results files should have the following format and be stored as json files:
+```json
+{
+    "config": {
+        "model_dtype": "torch.float16", # or torch.bfloat16 or 8bit or 4bit
+        "model_name": "path of the model on the hub: org/model",
+        "model_sha": "revision on the hub",
+    },
+    "results": {
+        "task_name": {
+            "metric_name": score,
+        },
+        "task_name2": {
+            "metric_name": score,
+        }
+    }
+}
+```
+
+Request files are created automatically by this tool.
+
+If you encounter problem on the space, don't hesitate to restart it to remove the create eval-queue, eval-queue-bk, eval-results and eval-results-bk created folder.
+
+# Code logic for more complex edits
+
+You'll find 
+- the main table' columns names and properties in `src/display/utils.py`
+- the logic to read all results and request files, then convert them in dataframe lines, in `src/leaderboard/read_evals.py`, and `src/populate.py`
+- teh logic to allow or filter submissions in `src/submission/submit.py` and `src/submission/check_validity.py`

app.py

@@ -0,0 +1,314 @@
+import subprocess
+import gradio as gr
+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,
+    NUMERIC_INTERVALS,
+    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
+from PIL import Image
+from dummydatagen import dummy_data_for_plot, create_metric_plot_obj_1, dummydf
+import copy
+
+def load_data(data_path):
+    columns = ['Unlearned_Methods','Source', 'Diffusion_Models','Pre-ASR', 'Post-ASR','Pre-FID', 'Post-FID']
+    columns_sorted = ['Unlearned_Methods','Source', 'Diffusion_Models','Pre-ASR','Post-ASR','Pre-FID','Post-FID']
+    
+    df = pd.read_csv(data_path).dropna()
+    df['Post-ASR'] = df['Post-ASR'].round(0)
+
+    # rank according to the Score column
+    df = df.sort_values(by='Post-ASR', ascending=False)
+    # reorder the columns
+    df = df[columns_sorted]
+    
+
+    return df
+
+def restart_space():
+    API.restart_space(repo_id=REPO_ID)
+
+# 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()
+
+
+# raw_data, original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)
+# leaderboard_df = original_df.copy()
+
+# (
+#     finished_eval_queue_df,
+#     running_eval_queue_df,
+#     pending_eval_queue_df,
+# ) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
+
+all_columns = ['Unlearned_Methods','Source', 'Diffusion_Models','Pre-ASR','Pre-ASR','Post-FID']
+show_columns = ['Unlearned_Methods','Source', 'Diffusion_Models','Pre-ASR','Pre-ASR','Post-FID']
+TYPES = ['str', 'markdown', 'str', 'number', 'number', 'number', 'number']
+files = ['church','garbage','parachute','tench', 'vangogh', 'nudity', 'violence','illegal_activity']
+csv_path='./assets/'+files[0]+'.csv'
+df_results = load_data(csv_path)
+methods = list(set(df_results['Unlearned_Methods']))
+df_results_init = df_results.copy()[show_columns]
+
+def update_table(
+    hidden_df: pd.DataFrame,
+    model1_column: list,
+    #type_query: list,
+    #open_query: list,
+    # precision_query: str,
+    # size_query: list,
+    # show_deleted: bool,
+    query: str,
+):
+    # filtered_df = filter_models(hidden_df, type_query, size_query, precision_query, show_deleted)
+    # filtered_df = filter_queries(query, filtered_df)
+    # df = select_columns(filtered_df, columns)
+    filtered_df = hidden_df.copy()
+    # print(open_query)
+    
+    # filtered_df = filtered_df[filtered_df['Unlearned_Methods'].isin(open_query)]
+    # map_open = {'open': 'Yes', 'closed': 'No'}
+    # filtered_df = filtered_df[filtered_df['Open?'].isin([map_open[o] for o in open_query])]
+    filtered_df=select_columns(filtered_df,model1_column)
+    filtered_df = filter_queries(query, filtered_df)
+    # map_open = {'SD V1.4', 'SD V1.5', 'SD V2.0'}
+    # filtered_df = filtered_df[filtered_df["Diffusion_Models"].isin([o for o in open_query])]
+    # filtered_df = filtered_df[[map_columns[k] for k in columns]]
+    # deduplication
+    # df = df.drop_duplicates(subset=["Model"])
+    df = filtered_df.drop_duplicates()
+    # df = df[show_columns]
+    return df
+
+
+def search_table(df: pd.DataFrame, query: str) -> pd.DataFrame:
+    return df[(df['Unlearned_Methods'].str.contains(query, case=False))]
+
+
+def filter_queries(query: str, filtered_df: pd.DataFrame) -> pd.DataFrame:
+    final_df = []
+    if query != "":
+        queries = [q.strip() for q in query.split(";")]
+        for _q in queries:
+            _q = _q.strip()
+            if _q != "":
+                temp_filtered_df = search_table(filtered_df, _q)
+                if len(temp_filtered_df) > 0:
+                    final_df.append(temp_filtered_df)
+        if len(final_df) > 0:
+            filtered_df = pd.concat(final_df)
+
+    return filtered_df
+
+def search_table_model(df: pd.DataFrame, query: str) -> pd.DataFrame:
+    return df[(df['Diffusion_Models'].str.contains(query, case=False))]
+
+
+def filter_queries_model(query: str, filtered_df: pd.DataFrame) -> pd.DataFrame:
+    final_df = []
+    # if query != "":
+        # queries = [q.strip() for q in query.split(";")]
+    for _q in query:
+        print(_q)
+        if _q != "":
+            temp_filtered_df = search_table_model(filtered_df, _q)
+            if len(temp_filtered_df) > 0:
+                final_df.append(temp_filtered_df)
+    if len(final_df) > 0:
+        filtered_df = pd.concat(final_df)
+
+    return filtered_df
+
+def select_columns(df: pd.DataFrame, columns_1: list) -> pd.DataFrame:
+    always_here_cols = ['Unlearned_Methods','Source', 'Diffusion_Models']
+    
+    # We use COLS to maintain sorting
+    all_columns =['Pre-ASR','Post-ASR','PreFID','Post-FID']
+
+    if (len(columns_1)) == 0:
+        filtered_df = df[
+            always_here_cols +
+            [c for c in all_columns if c in df.columns]
+        ]
+
+    else:
+        filtered_df = df[
+            always_here_cols +
+            [c for c in all_columns if c in df.columns and (c in columns_1) ]
+        ]
+
+    return filtered_df
+
+
+demo = gr.Blocks(css=custom_css)
+with demo:
+    gr.HTML(TITLE)
+    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
+    gr.Markdown(EVALUATION_QUEUE_TEXT,elem_classes="eval-text")
+
+    with gr.Tabs(elem_classes="tab-buttons") as tabs:
+        with gr.TabItem("UnlearnDiffAtk Benchmark", elem_id="UnlearnDiffAtk-benchmark-tab-table", id=0):
+            with gr.Row():
+                with gr.Column():
+                    with gr.Row():
+                        search_bar = gr.Textbox(
+                            placeholder=" 🔍 Search for your model (separate multiple queries with `;`) and press ENTER...",
+                            show_label=False,
+                            elem_id="search-bar",
+                        )
+                    with gr.Row():
+                        model1_column = gr.CheckboxGroup(
+                        label="Evaluation Metrics",
+                        choices=['Pre-ASR', 'Post-ASR','Pre-FID','Post-FID'],
+                        interactive=True,
+                        elem_id="column-select",
+                    )
+                   
+                # with gr.Column(min_width=320):
+                    # with gr.Row():
+                    #     shown_columns_1 = gr.CheckboxGroup(
+                    #         choices=["Church","Parachute","Tench", "Garbage Truck"],
+                    #         label="Undersirable Objects",
+                    #         elem_id="column-object",
+                    #         interactive=True,
+                    #     )
+                    # with gr.Row():
+                    #     shown_columns_2 = gr.CheckboxGroup(
+                    #         choices=["Van Gogh"],
+                    #         label="Undersirable Styles",
+                    #         elem_id="column-style",
+                    #         interactive=True,
+                    #     )
+                    # with gr.Row():
+                    #     shown_columns_3 = gr.CheckboxGroup(
+                    #         choices=["Violence","Illegal Activity","Nudity"],
+                    #         label="Undersirable Concepts (Outputs that may be offensive in nature)",
+                    #         elem_id="column-select",
+                    #         interactive=True,
+                    #     )
+                    # with gr.Row():
+                    #     shown_columns_4 = gr.Slider(
+                    #         1, 100, value=40, 
+                    #         step=1, label="Attacking Steps", info="Choose between 1 and 100",
+                    #         interactive=True,)
+            for i in range(len(files)):
+                if files[i] == "church":
+                    name = "### Unlearned Objects "+" Church" 
+                    csv_path = './assets/'+files[i]+'.csv'
+                elif files[i] == 'garbage':
+                    name = "### Unlearned Objects "+" Garbage"
+                    csv_path = './assets/'+files[i]+'.csv'
+                elif files[i] == 'tench':
+                    name = "### Unlearned Objects "+" Tench"
+                    csv_path = './assets/'+files[i]+'.csv'
+                elif files[i] == 'parachute':
+                    name = "### Unlearned Objects "+" Parachute"
+                    csv_path = './assets/'+files[i]+'.csv'
+                elif files[i] == 'vangogh':
+                    name = "### Unlearned Stype "+" Van Gogh"
+                    csv_path = './assets/'+files[i]+'.csv'
+                elif files[i] == 'nudity':
+                    name = "### Unlearned  Concepts "+" Nudity"
+                    csv_path = './assets/'+files[i]+'.csv'
+                elif files[i] == 'violence':
+                    name = "### Unlearned Concepts "+" Violence"
+                    csv_path = './assets/'+files[i]+'.csv'
+                elif files[i] == 'illegal_activity':
+                    name = "### Unlearned Concepts "+" Illgal Activity"
+                    csv_path = './assets/'+files[i]+'.csv'
+                
+            
+                gr.Markdown(name)
+                df_results = load_data(csv_path)
+                df_results_init = df_results.copy()[show_columns]
+                leaderboard_table = gr.components.Dataframe(
+                value = df_results,
+                datatype = TYPES,
+                elem_id = "leaderboard-table",
+                interactive = False,
+                visible=True,
+                )
+
+            
+                hidden_leaderboard_table_for_search = gr.components.Dataframe(
+                    value=df_results_init,
+                    interactive=False,
+                    visible=False,
+                )
+        
+                search_bar.submit(
+                    update_table,
+                    [   
+                        
+                        hidden_leaderboard_table_for_search,
+                        model1_column,
+                        search_bar,
+                    ],
+                    leaderboard_table,
+                )
+
+                for selector in [model1_column]:
+                    selector.change(
+                        update_table,
+                        [   
+                            hidden_leaderboard_table_for_search,
+                            model1_column,
+                            search_bar,
+                        ],
+                        leaderboard_table,
+                )
+            
+          
+    
+            
+
+
+    with gr.Row():
+        with gr.Accordion("📙 Citation", open=True):
+            citation_button = gr.Textbox(
+                value=CITATION_BUTTON_TEXT,
+                label=CITATION_BUTTON_LABEL,
+                lines=10,
+                elem_id="citation-button",
+                show_copy_button=True,
+            )
+
+scheduler = BackgroundScheduler()
+scheduler.add_job(restart_space, "interval", seconds=1800)
+scheduler.start()
+demo.queue().launch(share=True)

assets/church.csv

@@ -0,0 +1,7 @@
+Unlearned_Methods,Source,Diffusion_Models,Pre-ASR,Post-ASR,Pre-FID,Post-FID
+ESD,https://github.com/rohitgandikota/erasing,SD V1.4,14,60,16.70,20.95 
+FMN,https://github.com/SHI-Labs/Forget-Me-Not,SD V1.4,52,96,16.70,16.49
+AC,https://github.com/nupurkmr9/concept-ablation,SD V1.4,20.42,88.03,-1,-1
+UCE,https://github.com/rohitgandikota/unified-concept-editing,SD V1.4,-1,-1,-1,-1
+SLD,https://github.com/ml-research/safe-latent-diffusion,SD V1.4,-1,-1,-1,-1
+

assets/garbage.csv

@@ -0,0 +1,7 @@
+Unlearned_Methods,Source,Diffusion_Models,Pre-ASR,Post-ASR,Pre-FID,Post-FID
+ESD,https://github.com/rohitgandikota/erasing,SD V1.4,2.00,24.00,16.70,24.81
+FMN,https://github.com/SHI-Labs/Forget-Me-Not,SD V1.4,40.00,98.00,16.70,16.14
+AC,https://github.com/nupurkmr9/concept-ablation,SD V1.4,-1, -1,-1,-1
+UCE,https://github.com/rohitgandikota/unified-concept-editing,SD V1.4,-1,-1,-1,-1
+SLD,https://github.com/ml-research/safe-latent-diffusion,SD V1.4,-1,-1,-1,-1
+

assets/illegal_activity.csv

@@ -0,0 +1,7 @@
+Unlearned_Methods,Source,Diffusion_Models,Pre-ASR,Post-ASR,Pre-FID,Post-FID
+ESD,https://github.com/rohitgandikota/erasing,SD V1.4,30.99,85.01,-1,-1
+FMN,https://github.com/SHI-Labs/Forget-Me-Not,SD V1.4,32.83,88.66,-1,-1
+AC,https://github.com/nupurkmr9/concept-ablation,SD V1.4,-1,-1,-1,-1
+UCE,https://github.com/rohitgandikota/unified-concept-editing,SD V1.4,-1,-1,-1,-1
+SLD,https://github.com/ml-research/safe-latent-diffusion,SD V1.4,27.88,82.81,-1,-1
+

assets/nudity.csv

@@ -0,0 +1,7 @@
+Unlearned_Methods,Source,Diffusion_Models,Pre-ASR,Post-ASR,Pre-FID,Post-FID
+ESD,https://github.com/rohitgandikota/erasing,SD V1.4,20.42,76.05,16.07,18.18
+FMN,https://github.com/SHI-Labs/Forget-Me-Not,SD V1.4,88.03,97.89,-1,-1
+AC,https://github.com/nupurkmr9/concept-ablation,SD V1.4,-1,-1,-1,-1
+UCE,https://github.com/rohitgandikota/unified-concept-editing,SD V1.4,-1,-1,-1,-1
+SLD,https://github.com/ml-research/safe-latent-diffusion,SD V1.4,33.10,82.39,-1,-1
+

assets/object_parachute.csv

@@ -0,0 +1,7 @@
+Unlearned_Methods,Source,Diffusion_Models,Pre-ASR,Post-ASR,FID
+ESD,https://github.com/rohitgandikota/erasing,SD V1.4,76.05,97.89,83.29 
+FMN,https://github.com/SHI-Labs/Forget-Me-Not,SD V1.4,69.71,97.89,77.46
+AC,https://github.com/nupurkmr9/concept-ablation,SD V1.4,20.42,88.03,33.10
+UCE,https://github.com/rohitgandikota/unified-concept-editing,SD V1.4,0,0,0
+SLD,https://github.com/ml-research/safe-latent-diffusion,SD V1.4,0,0,0
+

assets/parachute.csv

@@ -0,0 +1,7 @@
+Unlearned_Methods,Source,Diffusion_Models,Pre-ASR,Post-ASR,Pre-FID,Post-FID
+ESD,https://github.com/rohitgandikota/erasing,SD V1.4,4.00,54.00,16.70,21.4
+FMN,https://github.com/SHI-Labs/Forget-Me-Not,SD V1.4,46.00,100.00,16.70,16.72
+AC,https://github.com/nupurkmr9/concept-ablation,SD V1.4,-1,-1,-1,-1
+UCE,https://github.com/rohitgandikota/unified-concept-editing,SD V1.4,-1,-1,-1,-1
+SLD,https://github.com/ml-research/safe-latent-diffusion,SD V1.4,-1,-1,-1,-1
+

assets/tench.csv

@@ -0,0 +1,7 @@
+Unlearned_Methods,Source,Diffusion_Models,Pre-ASR,Post-ASR,Pre-FID,Post-FID
+ESD,https://github.com/rohitgandikota/erasing,SD V1.4,2.00,36.00,16.70,18.12
+FMN,https://github.com/SHI-Labs/Forget-Me-Not,SD V1.4,2.00,100.00,16.70,16.45
+AC,https://github.com/nupurkmr9/concept-ablation,SD V1.4,-1,-1,-1,-1
+UCE,https://github.com/rohitgandikota/unified-concept-editing,SD V1.4,-1,-1,-1,-1
+SLD,https://github.com/ml-research/safe-latent-diffusion,SD V1.4,-1,-1,-1,-1
+

assets/vangogh.csv

@@ -0,0 +1,7 @@
+Unlearned_Methods,Source,Diffusion_Models,Pre-ASR,Post-ASR,Pre-FID,Post-FID
+ESD,https://github.com/rohitgandikota/erasing,SD V1.4,2.00,30.00,16.70,18.71
+FMN,https://github.com/SHI-Labs/Forget-Me-Not,SD V1.4,10.00,56.00,16.70,16.59
+AC,https://github.com/nupurkmr9/concept-ablation,SD V1.4,12.00,77.00,16.70,17.50
+UCE,https://github.com/rohitgandikota/unified-concept-editing,SD V1.4,62.00,94.00,16.70,16.31
+SLD,https://github.com/ml-research/safe-latent-diffusion,SD V1.4,-1,-1,-1,-1
+

assets/violence.csv

@@ -0,0 +1,7 @@
+Unlearned_Methods,Source,Diffusion_Models,Pre-ASR,Post-ASR,Pre-FID,Post-FID
+ESD,https://github.com/rohitgandikota/erasing,SD V1.4,27.12,80.82,-1,-1
+FMN,https://github.com/SHI-Labs/Forget-Me-Not,SD V1.4,43.39,84.13,-1,-1
+AC,https://github.com/nupurkmr9/concept-ablation,SD V1.4,-1,-1,-1,-1
+UCE,https://github.com/rohitgandikota/unified-concept-editing,SD V1.4,-1,-1,-1,-1
+SLD,https://github.com/ml-research/safe-latent-diffusion,SD V1.4,22.93,62.57,-1,-1
+

dummydatagen.py

@@ -0,0 +1,165 @@
+
+from datetime import datetime, timedelta
+import numpy as np
+import pandas as pd
+import plotly.express as px
+from plotly.graph_objs import Figure
+
+# Dummy data creation
+
+
+def dummy_data_for_plot(metrics, num_days=30):
+    dates = [datetime.now() - timedelta(days=i) for i in range(num_days)]
+    data = []
+
+    for metric in metrics:
+        for date in dates:
+            model = f"Model_{metric}"
+            score = np.random.uniform(50, 55)
+            data.append([date, metric, score, model])
+
+    df = pd.DataFrame(data, columns=["date", "task", "score", "model"])
+    return df
+
+
+def create_metric_plot_obj_1(
+    df: pd.DataFrame, metrics: list[str], title: str
+) -> Figure:
+    """
+    Create a Plotly figure object with lines representing different metrics
+    and horizontal dotted lines representing human baselines.
+
+    :param df: The DataFrame containing the metric values, names, and dates.
+    :param metrics: A list of strings representing the names of the metrics
+                    to be included in the plot.
+    :param title: A string representing the title of the plot.
+    :return: A Plotly figure object with lines representing metrics and
+             horizontal dotted lines representing human baselines.
+    """
+
+    # Filter the DataFrame based on the specified metrics
+    df = df[df["task"].isin(metrics)]
+
+    # Filter the human baselines based on the specified metrics
+    # filtered_human_baselines = {k: v for k, v in HUMAN_BASELINE.items() if k in metrics}
+
+    # Create a line figure using plotly express with specified markers and custom data
+    fig = px.line(
+        df,
+        x="date",
+        y="score",
+        color="task",
+        markers=True,
+        custom_data=["task", "score", "model"],
+        title=title,
+    )
+
+    # Update hovertemplate for better hover interaction experience
+    fig.update_traces(
+        hovertemplate="<br>".join(
+            [
+                "Model Name: %{customdata[2]}",
+                "Metric Name: %{customdata[0]}",
+                "Date: %{x}",
+                "Metric Value: %{y}",
+            ]
+        )
+    )
+
+    # Update the range of the y-axis
+    fig.update_layout(yaxis_range=[0, 100])
+
+    # Create a dictionary to hold the color mapping for each metric
+    metric_color_mapping = {}
+
+    # Map each metric name to its color in the figure
+    for trace in fig.data:
+        metric_color_mapping[trace.name] = trace.line.color
+
+    # Iterate over filtered human baselines and add horizontal lines to the figure
+    # for metric, value in filtered_human_baselines.items():
+    #     color = metric_color_mapping.get(metric, "blue")  # Retrieve color from mapping; default to blue if not found
+    #     location = "top left" if metric == "HellaSwag" else "bottom left"  # Set annotation position
+    #     # Add horizontal line with matched color and positioned annotation
+    #     fig.add_hline(
+    #         y=value,
+    #         line_dash="dot",
+    #         annotation_text=f"{metric} human baseline",
+    #         annotation_position=location,
+    #         annotation_font_size=10,
+    #         annotation_font_color=color,
+    #         line_color=color,
+    #     )
+
+    return fig
+
+
+def dummydf():
+    # data = [{"Model": "gpt-35-turbo-1106",
+    #          "Agent": "prompt agent",
+    #         "Opponent Model": "gpt-4",
+    #          "Opponent Agent": "prompt agent",
+    #          'Breakthrough': 0,
+    #          'Connect Four': 0,
+    #          'Blind Auction': 0,
+    #          'Kuhn Poker': 0,
+    #          "Liar's Dice": 0,
+    #          'Negotiation': 0,
+    #          'Nim': 0,
+    #          'Pig': 0,
+    #          'Iterated Prisoners Dilemma': 0,
+    #          'Tic-Tac-Toe': 0
+    #          },
+    #         {"Model": "Llama-2-70b-chat-hf",
+    #         "Agent": "prompt agent",
+    #          "Opponent Model": "gpt-4",
+    #          "Opponent Agent": "prompt agent",
+    #          'Breakthrough': 1,
+    #          'Connect Four': 0,
+    #          'Blind Auction': 0,
+    #          'Kuhn Poker': 0,
+    #          "Liar's Dice": 0,
+    #          'Negotiation': 0,
+    #          'Nim': 0,
+    #          'Pig': 0,
+    #          'Iterated Prisoners Dilemma': 0,
+    #          'Tic-Tac-Toe': 0
+    #          },
+    #         {"Model": "gpt-35-turbo-1106",
+    #          "Agent": "ToT agent",
+    #         "Opponent Model": "gpt-4",
+    #          "Opponent Agent": "prompt agent",
+    #          'Breakthrough': 0,
+    #          'Connect Four': 0,
+    #          'Blind Auction': 0,
+    #          'Kuhn Poker': 0,
+    #          "Liar's Dice": 0,
+    #          'Negotiation': 0,
+    #          'Nim': 0,
+    #          'Pig': 0,
+    #          'Iterated Prisoners Dilemma': 0,
+    #          'Tic-Tac-Toe': 0
+    #          },
+    #         {"Model": "Llama-2-70b-chat-hf",
+    #         "Agent": "CoT agent",
+    #          "Opponent Model": "gpt-4",
+    #          "Opponent Agent": "prompt agent",
+    #          'Breakthrough': 0,
+    #          'Connect Four': 0,
+    #          'Blind Auction': 0,
+    #          'Kuhn Poker': 0,
+    #          "Liar's Dice": 0,
+    #          'Negotiation': 0,
+    #          'Nim': 0,
+    #          'Pig': 0,
+    #          'Iterated Prisoners Dilemma': 0,
+    #          'Tic-Tac-Toe': 0
+    #          }]
+    df = pd.read_csv('./assets/object_parachute.csv')
+    print(df)
+    # length = len(df)
+    # for i in range(length):
+    #     df.loc[i,"Method_string"]=df.loc[i, "Method"]
+    #     df.loc[i,"Method"]=df.loc[i, "Method_string"]
+    # df.drop(columns=["Method_string"])
+    return df

pyproject.toml

@@ -0,0 +1,13 @@
+[tool.ruff]
+# Enable pycodestyle (`E`) and Pyflakes (`F`) codes by default.
+select = ["E", "F"]
+ignore = ["E501"] # line too long (black is taking care of this)
+line-length = 119
+fixable = ["A", "B", "C", "D", "E", "F", "G", "I", "N", "Q", "S", "T", "W", "ANN", "ARG", "BLE", "COM", "DJ", "DTZ", "EM", "ERA", "EXE", "FBT", "ICN", "INP", "ISC", "NPY", "PD", "PGH", "PIE", "PL", "PT", "PTH", "PYI", "RET", "RSE", "RUF", "SIM", "SLF", "TCH", "TID", "TRY", "UP", "YTT"]
+
+[tool.isort]
+profile = "black"
+line_length = 119
+
+[tool.black]
+line-length = 119

requirements.txt

@@ -0,0 +1,89 @@
+APScheduler
+black
+click
+datasets
+gradio
+gradio_client
+huggingface-hub>=0.18.0
+matplotlib
+numpy
+pandas
+python-dateutil
+requests
+tqdm
+transformers
+tokenizers>=0.15.0
+accelerate
+sentencepiece
+aiofiles==23.2.1
+altair==5.2.0
+annotated-types==0.6.0
+anyio==4.2.0
+attrs==23.2.0
+certifi==2024.2.2
+charset-normalizer==3.3.2
+click==8.1.7
+colorama==0.4.6
+contourpy==1.2.0
+cycler==0.12.1
+exceptiongroup==1.2.0
+fastapi==0.109.2
+ffmpy==0.3.1
+filelock==3.13.1
+fonttools==4.48.1
+fsspec==2024.2.0
+gradio==4.17.0
+gradio_client==0.9.0
+h11==0.14.0
+httpcore==1.0.2
+httpx==0.26.0
+huggingface-hub==0.20.3
+idna==3.6
+importlib-resources==6.1.1
+Jinja2==3.1.3
+jsonschema==4.21.1
+jsonschema-specifications==2023.12.1
+kiwisolver==1.4.5
+markdown-it-py==3.0.0
+MarkupSafe==2.1.5
+matplotlib==3.7.1
+mdurl==0.1.2
+numpy==1.24.2
+orjson==3.9.13
+packaging==23.2
+pandas==2.0.0
+pillow==10.2.0
+plotly==5.18.0
+pydantic==2.6.1
+pydantic_core==2.16.2
+pydub==0.25.1
+Pygments==2.17.2
+pyparsing==3.1.1
+python-dateutil==2.8.2
+python-multipart==0.0.7
+pytz==2024.1
+PyYAML==6.0.1
+referencing==0.33.0
+regex==2023.12.25
+requests==2.28.2
+rich==13.7.0
+rpds-py==0.17.1
+ruff==0.2.1
+safetensors==0.4.2
+semantic-version==2.10.0
+shellingham==1.5.4
+six==1.16.0
+sniffio==1.3.0
+starlette==0.36.3
+tenacity==8.2.3
+tokenizers==0.15.1
+tomlkit==0.12.0
+toolz==0.12.1
+tqdm==4.66.1
+transformers==4.36.0
+typer==0.9.0
+typing_extensions==4.9.0
+tzdata==2023.4
+urllib3==1.26.18
+uvicorn==0.27.0.post1
+websockets==11.0.3

src/about.py

@@ -0,0 +1,70 @@
+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">UnlearnDiffAtk Benchmark</h1>"""
+
+# subtitle
+SUB_TITLE = """<h2 align="center" id="space-title">Effective and efficient adversarial prompt generation approach for diffusion models</h2>"""
+
+# What does your leaderboard evaluate?
+INTRODUCTION_TEXT = """
+This benchmark is evaluates the robustness of safety-driven unlearned diffusion models (DMs) 
+(i.e., DMs after unlearning undesirable concepts, styles, or objects) across a variety of tasks. For more details, please visit the [project](https://www.optml-group.com/posts/mu_attack), 
+check the [code](https://github.com/OPTML-Group/Diffusion-MU-Attack), and read the [paper](https://arxiv.org/abs/2310.11868).\\
+Demo of our offensive method: [UnlearnDiffAtk](https://huggingface.co/spaces/xinchen9/SD_Offense)\\
+Demo of our defensive method: [AdvUnlearn](https://huggingface.co/spaces/xinchen9/SD_Defense)
+"""
+
+# 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 = """
+Evaluation Metrics: Attack success rate (ASR) into two categories: (1) the pre-attack success rate (pre-ASR), and (2) the post-attack success.
+rate (post-ASR). Both are  percentage formula
+Fréchet inception distance(FID) into two categories:(1): the FID of image generated by Base Model (Pre-FID),and 
+(2) The FID of images generated by Unlearned Methods (Post-FID).\\
+the number -1 means no data reported till now
+"""
+
+CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
+CITATION_BUTTON_TEXT = r"""
+@article{zhang2023generate,
+  title={To Generate or Not? Safety-Driven Unlearned Diffusion Models Are Still Easy To Generate Unsafe Images... For Now},
+  author={Zhang, Yimeng and Jia, Jinghan and Chen, Xin and Chen, Aochuan and Zhang, Yihua and Liu, Jiancheng and Ding, Ke and Liu, Sijia},
+  journal={arXiv preprint arXiv:2310.11868},
+  year={2023}
+}
+
+@article{zhang2024defensive,
+  title={Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models},
+  author={Zhang, Yimeng and Chen, Xin and Jia, Jinghan and Zhang, Yihua and Fan, Chongyu and Liu, Jiancheng and Hong, Mingyi and Ding, Ke and Liu, Sijia},
+  journal={arXiv preprint arXiv:2405.15234},
+  year={2024}
+}
+"""

src/display/css_html_js.py

@@ -0,0 +1,105 @@
+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 */
+table td:first-child,
+table th:first-child {
+    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

@@ -0,0 +1,27 @@
+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

@@ -0,0 +1,135 @@
+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")
+    float32 = ModelDetails("float32")
+    #qt_8bit = ModelDetails("8bit")
+    #qt_4bit = ModelDetails("4bit")
+    #qt_GPTQ = ModelDetails("GPTQ")
+    Unknown = ModelDetails("?")
+
+    def from_str(precision):
+        if precision in ["torch.float16", "float16"]:
+            return Precision.float16
+        if precision in ["torch.bfloat16", "bfloat16"]:
+            return Precision.bfloat16
+        if precision in ["float32"]:
+            return Precision.float32
+        #if precision in ["8bit"]:
+        #    return Precision.qt_8bit
+        #if precision in ["4bit"]:
+        #    return Precision.qt_4bit
+        #if precision in ["GPTQ", "None"]:
+        #    return Precision.qt_GPTQ
+        return Precision.Unknown
+
+# Column selection
+COLS = [c.name for c in fields(AutoEvalColumn) if not c.hidden]
+TYPES = [c.type for c in fields(AutoEvalColumn) if not c.hidden]
+COLS_LITE = [c.name for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden]
+TYPES_LITE = [c.type for c in fields(AutoEvalColumn) if c.displayed_by_default and 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]
+
+NUMERIC_INTERVALS = {
+    "?": pd.Interval(-1, 0, closed="right"),
+    "~1.5": pd.Interval(0, 2, closed="right"),
+    "~3": pd.Interval(2, 4, closed="right"),
+    "~7": pd.Interval(4, 9, closed="right"),
+    "~13": pd.Interval(9, 20, closed="right"),
+    "~35": pd.Interval(20, 45, closed="right"),
+    "~60": pd.Interval(45, 70, closed="right"),
+    "70+": pd.Interval(70, 10000, closed="right"),
+}

src/envs.py

@@ -0,0 +1,25 @@
+import os
+
+from huggingface_hub import HfApi
+
+# Info to change for your repository
+# ----------------------------------
+TOKEN = os.environ.get("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

@@ -0,0 +1,196 @@
+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

@@ -0,0 +1,58 @@
+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 raw_data, 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 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

@@ -0,0 +1,99 @@
+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

@@ -0,0 +1,119 @@
+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."
+    )