Spaces:
Sleeping
Sleeping
File size: 10,190 Bytes
b650828 3c6c62c b650828 3c6c62c b650828 3c6c62c b650828 3c6c62c 94bd921 3c6c62c b650828 9c9b7f5 12b667b 9c9b7f5 b650828 94bd921 3c6c62c 9c9b7f5 3c6c62c 94bd921 b650828 94bd921 b650828 94bd921 b650828 94bd921 b650828 94bd921 b650828 3c6c62c b650828 94bd921 9470d32 94bd921 b650828 01cbe38 b650828 3c6c62c 94bd921 3c6c62c b650828 94bd921 b650828 3c6c62c 94bd921 b650828 94bd921 b650828 94bd921 b650828 94bd921 b650828 94bd921 b650828 3c6c62c b650828 3c6c62c 70729a7 b650828 3c6c62c b650828 94bd921 b650828 3c6c62c b650828 3c6c62c b650828 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 |
import os
import json
import gradio as gr
import pandas as pd
import numpy as np
from pathlib import Path
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,
ABOUT_TEXT
)
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
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:
pass
# restart_space()
SUBSET_COUNTS = {
"Alignment-Object": 250,
"Alignment-Attribute": 229,
"Alignment-Action": 115,
"Alignment-Count": 55,
"Alignment-Location": 75,
"Safety-Toxicity-Crime": 29,
"Safety-Toxicity-Shocking": 31,
"Safety-Toxicity-Disgust": 42,
"Safety-Nsfw-Evident": 197,
"Safety-Nsfw-Evasive": 177,
"Safety-Nsfw-Subtle": 98,
"Quality-Distortion-Human_face": 169,
"Quality-Distortion-Human_limb": 152,
"Quality-Distortion-Object": 100,
"Quality-Blurry-Defocused": 350,
"Quality-Blurry-Motion": 350,
"Bias-Age": 80,
"Bias-Gender": 140,
"Bias-Race": 140,
"Bias-Nationality": 120,
"Bias-Religion": 60,
}
PERSPECTIVE_COUNTS= {
"Alignment": 724,
"Safety": 574,
"Quality": 1121,
"Bias": 540
}
META_DATA = ['Model']
def restart_space():
API.restart_space(repo_id=REPO_ID)
# color_map = {
# "Score Model": "#7497db",
# "Opensource VLM": "#E8ECF2",
# "Closesource VLM": "#ffcd75",
# "Others": "#75809c",
# # #7497db #E8ECF2 #ffcd75 #75809c
# }
# def color_model_type_column(df, color_map):
# """
# Apply color to the 'Model Type' column of the DataFrame based on a given color mapping.
# Parameters:
# df (pd.DataFrame): The DataFrame containing the 'Model Type' column.
# color_map (dict): A dictionary mapping model types to colors.
# Returns:
# pd.Styler: The styled DataFrame.
# """
# # Function to apply color based on the model type
# def apply_color(val):
# color = color_map.get(val, "default") # Default color if not specified in color_map
# return f'background-color: {color}'
# # Format for different columns
# format_dict = {col: "{:.1f}" for col in df.columns if col not in META_DATA}
# format_dict['Overall Score'] = "{:.2f}"
# format_dict[''] = "{:d}"
# return df.style.applymap(apply_color, subset=['Model Type']).format(format_dict, na_rep='')
def regex_table(dataframe, regex, filter_button, style=True):
"""
Takes a model name as a regex, then returns only the rows that has that in it.
"""
# Split regex statement by comma and trim whitespace around regexes
regex_list = [x.strip() for x in regex.split(",")]
# Join the list into a single regex pattern with '|' acting as OR
combined_regex = '|'.join(regex_list)
# if filter_button, remove all rows with "ai2" in the model name
update_scores = False
if isinstance(filter_button, list) or isinstance(filter_button, str):
if "Integrated LVLM" not in filter_button:
dataframe = dataframe[~dataframe["Model Type"].str.contains("Integrated LVLM", case=False, na=False)]
if "Interleaved LVLM" not in filter_button:
dataframe = dataframe[~dataframe["Model Type"].str.contains("Interleaved LVLM", case=False, na=False)]
# Filter the dataframe such that 'model' contains any of the regex patterns
data = dataframe[dataframe["Model"].str.contains(combined_regex, case=False, na=False)]
data.reset_index(drop=True, inplace=True)
# replace column '' with count/rank
data.insert(0, '', range(1, 1 + len(data)))
# if style:
# # apply color
# data = color_model_type_column(data, color_map)
return data
def get_leaderboard_results(results_path):
data_dir = Path(results_path)
files = [d for d in os.listdir(data_dir)] # TODO check if "Path(data_dir) / d" is a dir
df = pd.DataFrame()
for file in files:
if not file.endswith(".json"):
continue
with open(results_path / file) as rf:
result = json.load(rf)
result = pd.DataFrame(result)
df = pd.concat([result, df])
df.reset_index(drop=True, inplace=True)
return df
def avg_all_perspective(orig_df: pd.DataFrame, columns_name: list, meta_data=META_DATA, perspective_counts=PERSPECTIVE_COUNTS):
new_df = orig_df[meta_data + columns_name]
new_perspective_counts = {col: perspective_counts[col] for col in columns_name}
total_count = sum(perspective_counts.values())
weights = {perspective: count / total_count for perspective, count in perspective_counts.items()}
def calculate_weighted_avg(row):
weighted_sum = sum(row[col] * weights[col] for col in columns_name)
return weighted_sum
new_df["Overall Score"] = new_df.apply(calculate_weighted_avg, axis=1)
cols = meta_data + ["Overall Score"] + columns_name
new_df = new_df[cols].sort_values(by="Overall Score", ascending=False).reset_index(drop=True)
return new_df
data = {
"Model": [
"MiniGPT-5", "EMU-2", "GILL", "Anole",
"GPT-4o | Openjourney", "GPT-4o | SD-3", "GPT-4o | SD-XL", "GPT-4o | Flux",
"Gemini-1.5 | Openjourney", "Gemini-1.5 | SD-3", "Gemini-1.5 | SD-XL", "Gemini-1.5 | Flux",
"LLAVA-34b | Openjourney", "LLAVA-34b | SD-3", "LLAVA-34b | SD-XL", "LLAVA-34b | Flux",
"Qwen-VL-70b | Openjourney", "Qwen-VL-70b | SD-3", "Qwen-VL-70b | SD-XL", "Qwen-VL-70b | Flux"
],
"Model Type":[
"Interleaved LVLM", "Interleaved LVLM", "Interleaved LVLM", "Interleaved LVLM",
"Integrated LVLM", "Integrated LVLM", "Integrated LVLM", "Integrated LVLM",
"Integrated LVLM", "Integrated LVLM", "Integrated LVLM", "Integrated LVLM",
"Integrated LVLM", "Integrated LVLM", "Integrated LVLM", "Integrated LVLM",
"Integrated LVLM", "Integrated LVLM", "Integrated LVLM", "Integrated LVLM",
],
"Situational analysis": [
47.63, 39.65, 46.72, 48.95,
53.05, 53.00, 56.12, 54.97,
48.08, 47.48, 49.43, 47.07,
54.12, 54.72, 55.97, 54.23,
52.73, 54.98, 52.58, 54.23
],
"Project-based learning": [
55.12, 46.12, 57.57, 59.05,
71.40, 71.20, 73.25, 68.80,
67.93, 68.70, 71.85, 68.33,
73.47, 72.55, 74.60, 71.32,
71.63, 71.87, 73.57, 69.47
],
"Multi-step reasoning": [
42.17, 50.75, 39.33, 51.72,
53.67, 53.67, 53.67, 53.67,
60.05, 60.05, 60.05, 60.05,
47.28, 47.28, 47.28, 47.28,
55.63, 55.63, 55.63, 55.63
],
"AVG": [
50.92, 45.33, 51.58, 55.22,
63.65, 63.52, 65.47, 62.63,
61.57, 61.87, 64.15, 61.55,
63.93, 63.57, 65.05, 62.73,
64.05, 64.75, 65.12, 63.18
]
}
df = pd.DataFrame(data)
total_models = len(df)
with gr.Blocks(css=custom_css) as app:
with gr.Row():
with gr.Column(scale=6):
gr.Markdown(INTRODUCTION_TEXT.format(str(total_models)))
with gr.Column(scale=4):
gr.Markdown("")
# gr.HTML(BGB_LOGO, elem_classes="logo")
with gr.Tabs(elem_classes="tab-buttons") as tabs:
with gr.TabItem("🏆 MMIE Leaderboard"):
with gr.Row():
search_overall = gr.Textbox(
label="Model Search (delimit with , )",
placeholder="🔍 Search model (separate multiple queries with ``) and press ENTER...",
show_label=False
)
model_type_overall = gr.CheckboxGroup(
choices=["Interleaved LVLM", "Integrated LVLM"],
value=["Interleaved LVLM", "Integrated LVLM"],
label="Model Type",
show_label=False,
interactive=True,
)
with gr.Row():
mmie_table_overall_hidden = gr.Dataframe(
df,
headers=df.columns.tolist(),
elem_id="mmie_leadboard_overall_hidden",
wrap=True,
visible=False,
)
mmie_table_overall = gr.Dataframe(
regex_table(
df.copy(),
"",
["Interleaved LVLM", "Integrated LVLM"]
),
headers=df.columns.tolist(),
elem_id="mmie_leadboard_overall",
wrap=True,
)
with gr.TabItem("About"):
with gr.Row():
gr.Markdown(ABOUT_TEXT)
with gr.Accordion("📚 Citation", open=False):
citation_button = gr.Textbox(
value=CITATION_BUTTON_TEXT,
lines=7,
label="Copy the following to cite these results.",
elem_id="citation-button",
show_copy_button=True,
)
search_overall.change(regex_table, inputs=[mmie_table_overall_hidden, search_overall, model_type_overall], outputs=mmie_table_overall)
model_type_overall.change(regex_table, inputs=[mmie_table_overall_hidden, search_overall, model_type_overall], outputs=mmie_table_overall)
scheduler = BackgroundScheduler()
scheduler.add_job(restart_space, "interval", seconds=18000) # restarted every 3h
scheduler.start()
# app.queue(default_concurrency_limit=40).launch()
app.launch(allowed_paths=['./', "./src", "./evals"]) |