Spaces:
Sleeping
Sleeping
File size: 16,111 Bytes
3573a39 8f809e2 73ca636 3573a39 8f809e2 3573a39 53fe897 8f809e2 ed3fe33 b56bfdc ed3fe33 53fe897 5f9a95f 35be7f4 7f86019 53fe897 1b20cb8 7f86019 53fe897 5f9a95f 0607989 666860b 0607989 53fe897 0607989 8e32a09 3573a39 5b8d6d5 8f809e2 35be7f4 0fe497d 8c47a22 5f9a95f 8c47a22 35be7f4 8c47a22 666860b 35be7f4 0fe497d 61fc9c6 a89f9d8 53fe897 8f809e2 666860b 3573a39 35be7f4 8f809e2 0607989 35be7f4 6040976 35be7f4 666860b 35be7f4 53fe897 35be7f4 6040976 0607989 6040976 35be7f4 6040976 35be7f4 53fe897 8f809e2 cdcc48e 58c39e0 f25dac2 136af2d 5b8d6d5 8f809e2 5b8d6d5 f25dac2 8f809e2 136af2d 8f809e2 5b8d6d5 7487fdb 5b8d6d5 f25dac2 5b8d6d5 f25dac2 35be7f4 5b8d6d5 fdffa47 5b8d6d5 7487fdb 5b8d6d5 3573a39 35be7f4 3573a39 5b8d6d5 3573a39 8f809e2 5b8d6d5 8f809e2 3573a39 5b8d6d5 8f809e2 f25dac2 666860b 5f9a95f 35be7f4 7f86019 666860b 35be7f4 7055d8b 5311dba 7055d8b 35be7f4 0607989 35be7f4 fdffa47 7055d8b 5311dba 7055d8b 35be7f4 7055d8b 5311dba 7055d8b 35be7f4 7055d8b 5311dba 7055d8b 35be7f4 0607989 7055d8b 5311dba 7055d8b 35be7f4 5b8d6d5 ed3fe33 3573a39 5f9a95f 35be7f4 5f9a95f 8f809e2 5b8d6d5 5559b52 3573a39 8f809e2 3573a39 0607989 8e32a09 3573a39 7f86019 0607989 3573a39 7f86019 5311dba 7f86019 5311dba 7f86019 35be7f4 0607989 fdffa47 35be7f4 8f809e2 1c00552 8f809e2 5311dba 8f809e2 5b8d6d5 5559b52 3573a39 8f809e2 0607989 5311dba 0607989 3573a39 8f809e2 35be7f4 5b8d6d5 8f809e2 3573a39 35be7f4 3573a39 8f809e2 1c00552 2694247 8f809e2 4a799a0 5559b52 3573a39 8f809e2 666860b 2694247 5311dba 8f809e2 4a799a0 5559b52 3573a39 8f809e2 f25dac2 5b8d6d5 8f809e2 b56bfdc 8f809e2 cdcc48e 8f809e2 b56bfdc cdcc48e 5b8d6d5 4a799a0 cdcc48e 44e6352 cdcc48e f25dac2 fdffa47 afd881d 7487fdb fdffa47 7487fdb fdffa47 7487fdb 8217e92 3573a39 8f809e2 b56bfdc 8f809e2 fdffa47 8f809e2 fdffa47 5b8d6d5 1b20cb8 4a799a0 5b8d6d5 cdcc48e 7487fdb 0607989 fdffa47 8f809e2 5b8d6d5 ed3fe33 5b8d6d5 ed3fe33 5b8d6d5 5f9a95f 8f809e2 b56bfdc 5b8d6d5 73ca636 0607989 b56bfdc 5b8d6d5 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 |
import collections
import logging
import threading
import uuid
import datasets
import gradio as gr
import pandas as pd
import leaderboard
from io_utils import (
read_column_mapping,
write_column_mapping,
read_scanners,
write_scanners,
)
from run_jobs import save_job_to_pipe
from text_classification import (
strip_model_id_from_url,
check_model_task,
preload_hf_inference_api,
get_example_prediction,
get_labels_and_features_from_dataset,
check_hf_token_validity,
HuggingFaceInferenceAPIResponse,
)
from wordings import (
CHECK_CONFIG_OR_SPLIT_RAW,
CONFIRM_MAPPING_DETAILS_FAIL_RAW,
MAPPING_STYLED_ERROR_WARNING,
NOT_TEXT_CLASSIFICATION_MODEL_RAW,
UNMATCHED_MODEL_DATASET_STYLED_ERROR,
CHECK_LOG_SECTION_RAW,
VALIDATED_MODEL_DATASET_STYLED,
get_dataset_fetch_error_raw,
)
import os
from app_env import HF_WRITE_TOKEN
MAX_LABELS = 40
MAX_FEATURES = 20
ds_dict = None
ds_config = None
def get_related_datasets_from_leaderboard(model_id, dataset_id_input):
records = leaderboard.records
model_id = strip_model_id_from_url(model_id)
model_records = records[records["model_id"] == model_id]
datasets_unique = list(model_records["dataset_id"].unique())
if len(datasets_unique) == 0:
return gr.update(choices=[])
if dataset_id_input in datasets_unique:
return gr.update(choices=datasets_unique)
return gr.update(choices=datasets_unique, value="")
logger = logging.getLogger(__file__)
def get_dataset_splits(dataset_id, dataset_config):
try:
splits = datasets.get_dataset_split_names(dataset_id, dataset_config, trust_remote_code=True)
return gr.update(choices=splits, value=splits[0], visible=True)
except Exception as e:
logger.warn(f"Check your dataset {dataset_id} and config {dataset_config}: {e}")
return gr.update(visible=False)
def check_dataset(dataset_id):
logger.info(f"Loading {dataset_id}")
try:
configs = datasets.get_dataset_config_names(dataset_id, trust_remote_code=True)
if len(configs) == 0:
return (
gr.update(visible=False),
gr.update(visible=False),
""
)
splits = datasets.get_dataset_split_names(dataset_id, configs[0], trust_remote_code=True)
return (
gr.update(choices=configs, value=configs[0], visible=True),
gr.update(choices=splits, value=splits[0], visible=True),
""
)
except Exception as e:
logger.warn(f"Check your dataset {dataset_id}: {e}")
if "doesn't exist" in str(e):
gr.Warning(get_dataset_fetch_error_raw(e))
if "forbidden" in str(e).lower(): # GSK-2770
gr.Warning(get_dataset_fetch_error_raw(e))
return (
gr.update(visible=False),
gr.update(visible=False),
""
)
def empty_column_mapping(uid):
write_column_mapping(None, uid)
def write_column_mapping_to_config(uid, *labels):
# TODO: Substitute 'text' with more features for zero-shot
# we are not using ds features because we only support "text" for now
all_mappings = read_column_mapping(uid)
if labels is None:
return
all_mappings = export_mappings(all_mappings, "labels", None, labels[:MAX_LABELS])
all_mappings = export_mappings(
all_mappings,
"features",
["text"],
labels[MAX_LABELS : (MAX_LABELS + MAX_FEATURES)],
)
write_column_mapping(all_mappings, uid)
def export_mappings(all_mappings, key, subkeys, values):
if key not in all_mappings.keys():
all_mappings[key] = dict()
if subkeys is None:
subkeys = list(all_mappings[key].keys())
if not subkeys:
logging.debug(f"subkeys is empty for {key}")
return all_mappings
for i, subkey in enumerate(subkeys):
if subkey:
all_mappings[key][subkey] = values[i % len(values)]
return all_mappings
def list_labels_and_features_from_dataset(ds_labels, ds_features, model_labels, uid):
all_mappings = read_column_mapping(uid)
# For flattened raw datasets with no labels
# check if there are shared labels between model and dataset
shared_labels = set(model_labels).intersection(set(ds_labels))
if shared_labels:
ds_labels = list(shared_labels)
if len(ds_labels) > MAX_LABELS:
ds_labels = ds_labels[:MAX_LABELS]
gr.Warning(f"Too many labels to display for this spcae. We do not support more than {MAX_LABELS} in this space. You can use cli tool at https://github.com/Giskard-AI/cicd.")
# sort labels to make sure the order is consistent
# prediction gives the order based on probability
ds_labels.sort()
model_labels.sort()
lables = [
gr.Dropdown(
label=f"{label}",
choices=model_labels,
value=model_labels[i % len(model_labels)],
interactive=True,
visible=True,
)
for i, label in enumerate(ds_labels)
]
lables += [gr.Dropdown(visible=False) for _ in range(MAX_LABELS - len(lables))]
all_mappings = export_mappings(all_mappings, "labels", ds_labels, model_labels)
# TODO: Substitute 'text' with more features for zero-shot
features = [
gr.Dropdown(
label=f"{feature}",
choices=ds_features,
value=ds_features[0],
interactive=True,
visible=True,
)
for feature in ["text"]
]
features += [
gr.Dropdown(visible=False) for _ in range(MAX_FEATURES - len(features))
]
all_mappings = export_mappings(all_mappings, "features", ["text"], ds_features)
write_column_mapping(all_mappings, uid)
return lables + features
def precheck_model_ds_enable_example_btn(
model_id, dataset_id, dataset_config, dataset_split
):
model_id = strip_model_id_from_url(model_id)
model_task = check_model_task(model_id)
preload_hf_inference_api(model_id)
if dataset_config is None or dataset_split is None or len(dataset_config) == 0:
return (
gr.update(interactive=False),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
)
try:
ds = datasets.load_dataset(dataset_id, dataset_config, trust_remote_code=True)
df: pd.DataFrame = ds[dataset_split].to_pandas().head(5)
ds_labels, ds_features, _ = get_labels_and_features_from_dataset(ds[dataset_split])
if model_task is None or model_task != "text-classification":
gr.Warning(NOT_TEXT_CLASSIFICATION_MODEL_RAW)
return (
gr.update(interactive=False),
gr.update(value=df, visible=True),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
)
if not isinstance(ds_labels, list) or not isinstance(ds_features, list):
gr.Warning(CHECK_CONFIG_OR_SPLIT_RAW)
return (
gr.update(interactive=False),
gr.update(value=df, visible=True),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
)
return (
gr.update(interactive=True),
gr.update(value=df, visible=True),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
)
except Exception as e:
# Config or split wrong
logger.warn(f"Check your dataset {dataset_id} and config {dataset_config} on split {dataset_split}: {e}")
return (
gr.update(interactive=False),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
)
def align_columns_and_show_prediction(
model_id,
dataset_id,
dataset_config,
dataset_split,
uid,
inference_token,
):
model_id = strip_model_id_from_url(model_id)
model_task = check_model_task(model_id)
if model_task is None or model_task != "text-classification":
gr.Warning(NOT_TEXT_CLASSIFICATION_MODEL_RAW)
return (
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False, open=False),
gr.update(interactive=False),
"",
*[gr.update(visible=False) for _ in range(MAX_LABELS + MAX_FEATURES)],
)
dropdown_placement = [
gr.Dropdown(visible=False) for _ in range(MAX_LABELS + MAX_FEATURES)
]
hf_token = os.environ.get(HF_WRITE_TOKEN, default="")
prediction_input, prediction_response = get_example_prediction(
model_id, dataset_id, dataset_config, dataset_split, hf_token
)
if prediction_input is None or prediction_response is None:
return (
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False, open=False),
gr.update(interactive=False),
"",
*dropdown_placement,
)
if isinstance(prediction_response, HuggingFaceInferenceAPIResponse):
return (
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False, open=False),
gr.update(interactive=False),
f"Hugging Face Inference API is loading your model. {prediction_response.message}",
*dropdown_placement,
)
model_labels = list(prediction_response.keys())
ds = datasets.load_dataset(dataset_id, dataset_config, split=dataset_split, trust_remote_code=True)
ds_labels, ds_features, _ = get_labels_and_features_from_dataset(ds)
# when dataset does not have labels or features
if not isinstance(ds_labels, list) or not isinstance(ds_features, list):
gr.Warning(CHECK_CONFIG_OR_SPLIT_RAW)
return (
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False, open=False),
gr.update(interactive=False),
"",
*dropdown_placement,
)
if len(ds_labels) != len(model_labels):
return (
gr.update(value=UNMATCHED_MODEL_DATASET_STYLED_ERROR, visible=True),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False, open=False),
gr.update(interactive=False),
"",
*dropdown_placement,
)
column_mappings = list_labels_and_features_from_dataset(
ds_labels,
ds_features,
model_labels,
uid,
)
# when labels or features are not aligned
# show manually column mapping
if (
collections.Counter(model_labels) != collections.Counter(ds_labels)
or ds_features[0] != "text"
):
return (
gr.update(value=MAPPING_STYLED_ERROR_WARNING, visible=True),
gr.update(value=prediction_input, lines=min(len(prediction_input)//225 + 1, 5), visible=True),
gr.update(value=prediction_response, visible=True),
gr.update(visible=True, open=True),
gr.update(interactive=(inference_token != "")),
"",
*column_mappings,
)
return (
gr.update(value=VALIDATED_MODEL_DATASET_STYLED, visible=True),
gr.update(value=prediction_input, lines=min(len(prediction_input)//225 + 1, 5), visible=True),
gr.update(value=prediction_response, visible=True),
gr.update(visible=True, open=False),
gr.update(interactive=(inference_token != "")),
"",
*column_mappings,
)
def check_column_mapping_keys_validity(all_mappings):
if all_mappings is None:
logger.warning("all_mapping is None")
gr.Warning(CONFIRM_MAPPING_DETAILS_FAIL_RAW)
return False
if "labels" not in all_mappings.keys():
logger.warning(f"Label mapping is not valid, all_mappings: {all_mappings}")
return False
return True
def enable_run_btn(uid, inference_token, model_id, dataset_id, dataset_config, dataset_split):
if inference_token == "":
logger.warn("Inference API is not enabled")
return gr.update(interactive=False)
if model_id == "" or dataset_id == "" or dataset_config == "" or dataset_split == "":
logger.warn("Model id or dataset id is not selected")
return gr.update(interactive=False)
all_mappings = read_column_mapping(uid)
if not check_column_mapping_keys_validity(all_mappings):
logger.warn("Column mapping is not valid")
return gr.update(interactive=False)
if not check_hf_token_validity(inference_token):
logger.warn("HF token is not valid")
return gr.update(interactive=False)
return gr.update(interactive=True)
def construct_label_and_feature_mapping(all_mappings, ds_labels, ds_features, label_keys=None):
label_mapping = {}
if len(all_mappings["labels"].keys()) != len(ds_labels):
logger.warn(f"""Label mapping corrupted: {CONFIRM_MAPPING_DETAILS_FAIL_RAW}.
\nall_mappings: {all_mappings}\nds_labels: {ds_labels}""")
if len(all_mappings["features"].keys()) != len(ds_features):
logger.warn(f"""Feature mapping corrupted: {CONFIRM_MAPPING_DETAILS_FAIL_RAW}.
\nall_mappings: {all_mappings}\nds_features: {ds_features}""")
for i, label in zip(range(len(ds_labels)), ds_labels):
# align the saved labels with dataset labels order
label_mapping.update({str(i): all_mappings["labels"][label]})
if "features" not in all_mappings.keys():
logger.warning("features not in all_mappings")
gr.Warning(CONFIRM_MAPPING_DETAILS_FAIL_RAW)
feature_mapping = all_mappings["features"]
if len(label_keys) > 0:
feature_mapping.update({"label": label_keys[0]})
return label_mapping, feature_mapping
def show_hf_token_info(token):
valid = check_hf_token_validity(token)
if not valid:
return gr.update(visible=True)
return gr.update(visible=False)
def try_submit(m_id, d_id, config, split, inference_token, uid):
all_mappings = read_column_mapping(uid)
if not check_column_mapping_keys_validity(all_mappings):
return (gr.update(interactive=True), gr.update(visible=False))
# get ds labels and features again for alignment
ds = datasets.load_dataset(d_id, config, split=split, trust_remote_code=True)
ds_labels, ds_features, label_keys = get_labels_and_features_from_dataset(ds)
label_mapping, feature_mapping = construct_label_and_feature_mapping(all_mappings, ds_labels, ds_features, label_keys)
eval_str = f"[{m_id}]<{d_id}({config}, {split} set)>"
save_job_to_pipe(
uid,
(
m_id,
d_id,
config,
split,
inference_token,
uid,
label_mapping,
feature_mapping,
),
eval_str,
threading.Lock(),
)
gr.Info("Your evaluation has been submitted")
new_uid = uuid.uuid4()
scanners = read_scanners(uid)
write_scanners(scanners, new_uid)
return (
gr.update(interactive=False), # Submit button
gr.update(value=f"{CHECK_LOG_SECTION_RAW}Your job id is: {uid}. ", lines=5, visible=True, interactive=False),
new_uid, # Allocate a new uuid
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
gr.update(visible=False),
)
|