Spaces:
Running
Running
File size: 9,561 Bytes
e4ba316 4e5edbb e4ba316 4e5edbb e4ba316 4e5edbb e4ba316 4e5edbb e4ba316 4e5edbb e4ba316 4e5edbb e4ba316 4e5edbb e4ba316 4e5edbb e4ba316 4e5edbb e4ba316 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 | import json
import shutil
import tempfile
import traceback
import uuid
import zipfile
from pathlib import Path
APP_ROOT = Path(__file__).resolve().parent
RUNTIME_ARCHIVE = APP_ROOT / "lite_runtime.zip"
def _extract_runtime():
with zipfile.ZipFile(RUNTIME_ARCHIVE) as archive:
for member in archive.infolist():
target = (APP_ROOT / member.filename).resolve()
if APP_ROOT not in target.parents and target != APP_ROOT:
raise RuntimeError("The evaluator runtime archive contains an unsafe path.")
archive.extractall(APP_ROOT)
_extract_runtime()
import gradio as gr
from chinatravel.ui import content
from chinatravel.ui.archive import SubmissionArchiveError, extract_submission_archive
from chinatravel.ui.leaderboard import build_leaderboards
from eval_exp import evaluate
EVALUATION_SPLITS = ("easy", "human")
LEADERBOARD_SPLITS = ("easy", "human", "human1000")
LANGUAGE_SPLITS = {"zh": EVALUATION_SPLITS, "en": EVALUATION_SPLITS}
DEFAULT_SPLIT = "easy"
OVERVIEW_URL = (
"https://huggingface.co/spaces/LAMDA-NeSy/ChinaTravel/resolve/"
"main/images/overview.png"
)
class Arguments:
def __init__(self, splits, result_dir, lang):
self.splits = splits
self.result_dir = result_dir
self.lang = lang
self.oracle_translation = True
def _upload_path(zip_file):
if isinstance(zip_file, dict):
return zip_file.get("path") or zip_file.get("name")
return getattr(zip_file, "name", zip_file)
def splits_for_language(lang):
return LANGUAGE_SPLITS.get(lang, ())
def update_evaluation_splits(language, current_split):
choices = list(splits_for_language(language))
value = current_split if current_split in choices else choices[0]
return gr.update(choices=choices, value=value)
def handle_submission(zip_file, dataset_choice, language_choice):
empty_progress = content.render_progress()
empty_metrics = content.render_metrics()
if zip_file is None:
yield (
content.render_status(
"error",
"Archive required",
"Upload a ZIP archive containing prediction JSON files.",
),
empty_progress,
empty_metrics,
None,
)
return
if dataset_choice not in EVALUATION_SPLITS:
yield (
content.render_status(
"error",
"Unsupported split",
"Browser evaluation is available for easy and human.",
),
empty_progress,
empty_metrics,
None,
)
return
if language_choice not in LANGUAGE_SPLITS:
yield (
content.render_status(
"error",
"Invalid language",
"Select either Chinese or English evaluation.",
),
empty_progress,
empty_metrics,
None,
)
return
job_dir = Path(tempfile.mkdtemp(prefix="chinatravel-lite-"))
submission_dir = job_dir / "submission"
try:
summary = extract_submission_archive(_upload_path(zip_file), submission_dir)
yield (
content.render_status(
"running",
"Evaluation started",
f"Validated {summary.json_files} prediction files.",
),
empty_progress,
empty_metrics,
None,
)
args = Arguments(dataset_choice, str(submission_dir), language_choice)
stage_values = {"schema": 0, "commonsense": 0, "logic": 0}
result = {}
for update in evaluate(args, result):
stage = update.get("stage", "")
if stage in stage_values:
stage_values[stage] = max(
0,
min(100, round(update.get("progress", 0))),
)
labels = {
"schema": "Checking output schema",
"commonsense": "Checking environment grounding",
"logic": "Checking hard constraints",
}
yield (
content.render_status(
"running",
labels[stage],
f"Split: {dataset_choice} · Language: {language_choice}",
),
content.render_progress(
stage_values["schema"],
stage_values["commonsense"],
stage_values["logic"],
),
empty_metrics,
None,
)
elif stage == "final":
result.update(update.get("result", {}))
output_path = job_dir / (
f"evaluation-{dataset_choice}-{language_choice}-"
f"{uuid.uuid4().hex[:8]}.json"
)
output_path.write_text(
json.dumps(result, ensure_ascii=False, indent=2),
encoding="utf-8",
)
matched = len(result.get("matched_uid", []))
unmatched = len(result.get("unmatched_uid", []))
yield (
content.render_status(
"complete",
"Evaluation complete",
f"Matched {matched} predictions; {unmatched} expected UIDs were missing.",
),
content.render_progress(100, 100, 100),
content.render_metrics(result),
str(output_path),
)
except SubmissionArchiveError as exc:
yield (
content.render_status("error", "Archive rejected", str(exc)),
empty_progress,
empty_metrics,
None,
)
except Exception as exc:
traceback.print_exc()
yield (
content.render_status("error", "Evaluation failed", str(exc)),
empty_progress,
empty_metrics,
None,
)
finally:
shutil.rmtree(submission_dir, ignore_errors=True)
leaderboards = build_leaderboards(LEADERBOARD_SPLITS)
with gr.Blocks(
title="ChinaTravel Benchmark",
theme=gr.themes.Base(
primary_hue=gr.themes.colors.red,
secondary_hue=gr.themes.colors.emerald,
neutral_hue=gr.themes.colors.gray,
),
css=content.CSS,
fill_width=True,
) as demo:
gr.HTML(content.HEADER_HTML)
with gr.Tabs(elem_id="primary-tabs"):
with gr.Tab("Leaderboard"):
gr.HTML(content.LEADERBOARD_HEADER_HTML)
with gr.Tabs(elem_id="track-tabs"):
with gr.Tab("Standard"), gr.Tabs(elem_classes=["ct-split-tabs"]):
for split in LEADERBOARD_SPLITS:
with gr.Tab(split):
gr.HTML(leaderboards[split]["standard"])
with gr.Tab("Oracle-assisted"), gr.Tabs(
elem_classes=["ct-split-tabs"]
):
for split in LEADERBOARD_SPLITS:
with gr.Tab(split):
gr.HTML(leaderboards[split]["oracle"])
gr.HTML(content.LEADERBOARD_NOTE_HTML)
with gr.Tab("Evaluate"):
gr.HTML(content.EVALUATOR_HEADER_HTML)
with gr.Row(elem_classes=["ct-workbench"]):
with gr.Column(scale=4, min_width=280):
dataset_choice = gr.Dropdown(
list(EVALUATION_SPLITS),
label="Evaluation split",
value=DEFAULT_SPLIT,
interactive=True,
)
language_choice = gr.Radio(
choices=[("Chinese", "zh"), ("English", "en")],
value="zh",
label="Evaluation language",
interactive=True,
)
language_choice.change(
update_evaluation_splits,
inputs=[language_choice, dataset_choice],
outputs=dataset_choice,
)
zip_input = gr.File(
label="Prediction archive",
file_types=[".zip"],
type="filepath",
)
submit_btn = gr.Button(
"Run evaluation",
variant="primary",
elem_classes=["ct-submit-button"],
)
gr.HTML(content.EVALUATOR_RESOURCES_HTML)
with gr.Column(scale=7, min_width=340):
status = gr.HTML(content.render_status())
progress = gr.HTML(content.render_progress())
metrics = gr.HTML(content.render_metrics())
result_file = gr.File(
label="Evaluation report",
interactive=False,
)
submit_btn.click(
handle_submission,
inputs=[zip_input, dataset_choice, language_choice],
outputs=[status, progress, metrics, result_file],
)
with gr.Tab("Benchmark"):
gr.HTML(content.ABOUT_HTML)
gr.Image(
value=OVERVIEW_URL,
show_label=False,
interactive=False,
elem_classes=["overview-image"],
)
gr.HTML(content.FOOTER_HTML)
demo.launch(show_error=False)
|