File size: 9,940 Bytes
3c856c0 30c3967 3c856c0 30c3967 3c856c0 30c3967 3c856c0 30c3967 3c856c0 30c3967 feaf833 30c3967 7357a15 30c3967 feaf833 30c3967 feaf833 30c3967 7357a15 30c3967 3c856c0 30c3967 3c856c0 30c3967 3c856c0 30c3967 3c856c0 30c3967 3c856c0 30c3967 27c9b8f 3c856c0 7357a15 3c856c0 30c3967 3c856c0 |
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 |
import gradio as gr
import os
import json
import pandas as pd
from huggingface_hub import HfApi, hf_hub_download
from datasets import load_dataset
import requests
TOKEN = os.environ.get("HF_TOKEN")
OWNER = os.environ.get("OWNER")
RESULTS_COMMUNITY = f"{OWNER}/benchmark_results"
api = HfApi()
URL = os.environ.get("URL")
def load_data(source):
if source == "core":
with open("data.json", "r") as f:
data = json.load(f)
else:
ds = load_dataset(RESULTS_COMMUNITY, split='train')
data = []
for entry in ds:
data.append(entry)
return data
def build_table(source):
data = load_data(source)
entries = []
if source == "core":
headers = ["Benchmark", "Category", "Pile-train Dirty (%)", "DCLM-baseline Dirty (%)", "CC-2025-05 Dirty (%)"]
else:
headers = ["Benchmark", "Contributor", "Pile-train Dirty (%)", "DCLM-baseline Dirty (%)", "CC-2025-05 Dirty (%)"]
html = """
<table id="benchmarkTable" style="border-collapse: collapse; width: 100%;">
<thead><tr>
"""
for col in headers:
html += f'<th style="border: 1px solid #ddd; padding: 8px; text-align: right;" onclick="sortTable(this)">{col} <span class="triangle"></span></th>'
html += '</tr></thead>\n<tbody>\n'
for entry in data:
name = entry.get("Benchmark", "")
url = entry.get("URL", "#")
hyperlink = f'<a href="{url}" target="_blank">{name}</a>'
row = {
"Benchmark": hyperlink,
"Pile-train Dirty (%)": entry.get("Pile Dirty", -1),
"DCLM-baseline Dirty (%)": entry.get("DCLM Dirty", -1),
"CC-2025-05 Dirty (%)": entry.get("CC202505 Dirty", -1),
}
if source == "core":
row["Category"] = entry.get("Category", "")
elif source == "community":
row["Contributor"] = entry.get("Contributor", "")
html += "<tr>"
for col in headers:
val = row.get(col, "")
if isinstance(val, float) and val >= 0:
val = f"{val:5.1f}"
elif isinstance(val, float):
val = "N/A"
html += f'<td style="border: 1px solid #ddd; padding: 8px; text-align: right;">{val}</td>'
html += "</tr>\n"
html += "</tbody></table>"
html += """
<script>
let sortDirection = {};
function sortTable(header) {
var table = document.getElementById("benchmarkTable");
var rows = Array.from(table.rows).slice(1);
var columnIndex = Array.from(header.parentNode.children).indexOf(header);
var isAscending = sortDirection[columnIndex] === 'ascending';
sortDirection[columnIndex] = isAscending ? 'descending' : 'ascending';
var allHeaders = header.parentNode.children;
Array.from(allHeaders).forEach(th => {
th.querySelector('.triangle').classList.remove('ascending', 'descending');
});
header.querySelector('.triangle').classList.add(sortDirection[columnIndex]);
rows.sort(function(rowA, rowB) {
var cellA = rowA.cells[columnIndex].innerText;
var cellB = rowB.cells[columnIndex].innerText;
if (isNaN(cellA)) {
return isAscending ? cellA.localeCompare(cellB) : cellB.localeCompare(cellA);
}
return isAscending ? parseFloat(cellA) - parseFloat(cellB) : parseFloat(cellB) - parseFloat(cellA);
});
for (var i = 0; i < rows.length; i++) {
table.appendChild(rows[i]);
}
}
</script>
"""
html += """
<style>
.triangle {
display: inline-block;
width: 0;
height: 0;
border-left: 5px solid transparent;
border-right: 5px solid transparent;
margin-left: 5px;
transition: transform 0.2s;
}
.ascending {
border-bottom: 5px solid #000;
}
.descending {
border-top: 5px solid #000;
}
</style>
"""
return html
def record_submission(benchmark_name, contributor, jsonl_file, hf_path, hf_split, field_name):
if not benchmark_name or not benchmark_name.strip():
return "β Please provide a benchmark name."
if not field_name or not field_name.strip():
return "β Please provide a field name."
has_jsonl = jsonl_file is not None
has_hf = hf_path and hf_path.strip()
if not has_jsonl and not has_hf:
return "β Please provide either a .jsonl file or a HuggingFace dataset path."
if has_jsonl:
try:
with open(jsonl_file.name, 'r', encoding='utf-8') as f:
line_count = 0
for line in f:
line_count += 1
if line_count > 5:
break
try:
entry = json.loads(line.strip())
if field_name.strip() not in entry:
available_fields = list(entry.keys())
return f"β Field '{field_name.strip()}' not found in JSONL file. Available fields: {', '.join(available_fields)}"
except json.JSONDecodeError as e:
return f"β Invalid JSON format in line {line_count}: {str(e)}"
if line_count == 0:
return "β The uploaded file is empty."
except Exception as e:
return f"β Error reading file: {str(e)}"
elif has_hf:
if not hf_split or not hf_split.strip():
return "β Please provide a dataset split for the HuggingFace dataset."
try:
dataset_info = load_dataset(hf_path.strip(), split=hf_split.strip(), streaming=True, trust_remote_code=True)
first_item = next(iter(dataset_info))
if field_name.strip() not in first_item:
available_fields = list(first_item.keys())
return f"β Field '{field_name.strip()}' not found in dataset. Available fields: {', '.join(available_fields)}"
except Exception as e:
return f"β Could not access HuggingFace dataset: {str(e)}"
try:
data = {
'name': benchmark_name.strip(),
'contributor': contributor.strip(),
'type': 'jsonl' if has_jsonl else 'hf',
'split': hf_split.strip() if has_hf else '',
'field_name': field_name.strip(),
'hf_path': hf_path.strip() if has_hf else ''
}
print(json.dumps(data))
files = {}
if has_jsonl:
files['file'] = (benchmark_name.strip() + '.jsonl', open(jsonl_file.name, 'rb'), 'application/json')
response = requests.post(f"{URL}/", data={"payload": json.dumps(data)}, files=files, timeout=30)
if files:
files['file'][1].close()
if response.status_code == 200:
result = response.json()
if result.get("status") == "success":
message = result.get('message', 'Submission successful!')
full_message = f"{message}"
return full_message
elif result.get("status") == "info":
return f"β {result.get('message', 'Submission already exists')}"
else:
return f"β {result.get('message', 'Unknown error occurred')}"
else:
return f"β Server error: {response.status_code} - {response.text}"
except Exception as e:
return f"β Error submitting benchmark: {str(e)}"
with gr.Blocks() as interface:
gr.Markdown("# π Benchmark Contamination Bulletin")
with gr.Tabs():
with gr.Tab(label="Bulletin"):
source_radio = gr.Radio(
choices=["core", "community"],
label="Select Benchmark Source",
value="core"
)
leaderboard_html = gr.HTML(build_table("core"))
def update_table(source):
return build_table(source)
source_radio.change(
fn=update_table,
inputs=source_radio,
outputs=leaderboard_html
)
refresh_button = gr.Button("Refresh")
refresh_button.click(
fn=update_table,
inputs=source_radio,
outputs=leaderboard_html
)
with gr.Tab(label="Add New Benchmarks"):
gr.Markdown("## Add Your Own Benchmarks for Contamination Checking")
with gr.Row():
benchmark_name_input = gr.Textbox(label="Benchmark Name")
contributor_input = gr.Textbox(label="Contributor")
with gr.Row():
jsonl_input = gr.File(label="Upload .jsonl File", file_types=[".jsonl"])
with gr.Column():
hf_path_input = gr.Textbox(label="HuggingFace Dataset Path")
hf_split_input = gr.Textbox(label="Dataset split (only if providing HuggingFace Dataset)", placeholder="e.g., validation, test")
field_name_input = gr.Textbox(label="Context or Question Field Name", placeholder="e.g., context, question, ...")
submit_button = gr.Button("Submit for Contamination Check")
result_output = gr.Textbox(label="Submission Status", interactive=False)
submit_button.click(
fn=record_submission,
inputs=[benchmark_name_input, contributor_input, jsonl_input, hf_path_input, hf_split_input, field_name_input],
outputs=result_output
)
interface.launch()
|