File size: 18,023 Bytes
e18c8b0 a6b53fb e18c8b0 61173ea 727d1ca 5bfd036 3373091 10cef3f 5bfd036 e18c8b0 a6b53fb e18c8b0 7d317a5 e18c8b0 05bf37a 5bfd036 e18c8b0 ca9549b e18c8b0 05bf37a a6b53fb 10cef3f e18c8b0 5bfd036 e18c8b0 d4ce695 e18c8b0 05bf37a 5bfd036 e18c8b0 5bfd036 e18c8b0 91df4d6 ca9549b e18c8b0 ca9549b e18c8b0 ca9549b e18c8b0 5bfd036 e18c8b0 61173ea e18c8b0 5bfd036 727d1ca 5bfd036 e18c8b0 3373091 ca9549b 5bfd036 a6b53fb ca9549b 727d1ca ca9549b 5bfd036 ca9549b 5bfd036 ca9549b e18c8b0 05bf37a 5bfd036 a6b53fb e18c8b0 a6b53fb e18c8b0 a6b53fb e18c8b0 a6b53fb e18c8b0 b3e5cf7 e18c8b0 5bfd036 e6b76e7 5bfd036 e6b76e7 e18c8b0 ca9549b e18c8b0 10cef3f e18c8b0 a6b53fb e18c8b0 ac95e09 ca9549b ac95e09 ca9549b e18c8b0 ca9549b e18c8b0 05bf37a e18c8b0 5bfd036 e18c8b0 5bfd036 e18c8b0 10cef3f e18c8b0 727d1ca e18c8b0 727d1ca 05bf37a 727d1ca 10cef3f 727d1ca e18c8b0 727d1ca e18c8b0 7bcd3fa a6b53fb e18c8b0 a6b53fb e18c8b0 7bcd3fa e18c8b0 a6b53fb e18c8b0 a6b53fb 7bcd3fa e18c8b0 a6b53fb |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 |
import logging
import time
from datetime import timedelta
from typing import Dict, List
import streamlit as st
from llm_guard.input_scanners.anonymize import default_entity_types
from llm_guard.input_scanners.code import SUPPORTED_LANGUAGES as SUPPORTED_CODE_LANGUAGES
from llm_guard.output_scanners import get_scanner_by_name
from llm_guard.output_scanners.bias import MatchType as BiasMatchType
from llm_guard.output_scanners.deanonymize import MatchingStrategy as DeanonymizeMatchingStrategy
from llm_guard.output_scanners.gibberish import MatchType as GibberishMatchType
from llm_guard.output_scanners.language import MatchType as LanguageMatchType
from llm_guard.output_scanners.toxicity import MatchType as ToxicityMatchType
from llm_guard.vault import Vault
from streamlit_tags import st_tags
logger = logging.getLogger("llm-guard-playground")
def init_settings() -> (List, Dict):
all_scanners = [
"BanCode",
"BanCompetitors",
"BanSubstrings",
"BanTopics",
"Bias",
"Code",
"Deanonymize",
"JSON",
"Language",
"LanguageSame",
"MaliciousURLs",
"NoRefusal",
"NoRefusalLight" "ReadingTime",
"FactualConsistency",
"Gibberish",
"Regex",
"Relevance",
"Sensitive",
"Sentiment",
"Toxicity",
"URLReachability",
]
st_enabled_scanners = st.sidebar.multiselect(
"Select scanners",
options=all_scanners,
default=all_scanners,
help="The list can be found here: https://llm-guard.com/output_scanners/bias/",
)
settings = {}
if "BanCode" in st_enabled_scanners:
st_bc_expander = st.sidebar.expander(
"Ban Code",
expanded=False,
)
with st_bc_expander:
st_bc_threshold = st.slider(
label="Threshold",
value=0.95,
min_value=0.0,
max_value=1.0,
step=0.05,
key="ban_code_threshold",
)
settings["BanCode"] = {"threshold": st_bc_threshold}
if "BanCompetitors" in st_enabled_scanners:
st_bc_expander = st.sidebar.expander(
"Ban Competitors",
expanded=False,
)
with st_bc_expander:
st_bc_competitors = st_tags(
label="List of competitors",
text="Type and press enter",
value=["openai", "anthropic", "deepmind", "google"],
suggestions=[],
maxtags=30,
key="bc_competitors",
)
st_bc_threshold = st.slider(
label="Threshold",
value=0.5,
min_value=0.0,
max_value=1.0,
step=0.05,
key="ban_competitors_threshold",
)
settings["BanCompetitors"] = {
"competitors": st_bc_competitors,
"threshold": st_bc_threshold,
}
if "BanSubstrings" in st_enabled_scanners:
st_bs_expander = st.sidebar.expander(
"Ban Substrings",
expanded=False,
)
with st_bs_expander:
st_bs_substrings = st.text_area(
"Enter substrings to ban (one per line)",
value="test\nhello\nworld\n",
height=200,
).split("\n")
st_bs_match_type = st.selectbox(
"Match type", ["str", "word"], index=0, key="bs_match_type"
)
st_bs_case_sensitive = st.checkbox(
"Case sensitive", value=False, key="bs_case_sensitive"
)
st_bs_redact = st.checkbox("Redact", value=False, key="bs_redact")
st_bs_contains_all = st.checkbox("Contains all", value=False, key="bs_contains_all")
settings["BanSubstrings"] = {
"substrings": st_bs_substrings,
"match_type": st_bs_match_type,
"case_sensitive": st_bs_case_sensitive,
"redact": st_bs_redact,
"contains_all": st_bs_contains_all,
}
if "BanTopics" in st_enabled_scanners:
st_bt_expander = st.sidebar.expander(
"Ban Topics",
expanded=False,
)
with st_bt_expander:
st_bt_topics = st_tags(
label="List of topics",
text="Type and press enter",
value=["violence"],
suggestions=[],
maxtags=30,
key="bt_topics",
)
st_bt_threshold = st.slider(
label="Threshold",
value=0.6,
min_value=0.0,
max_value=1.0,
step=0.05,
key="ban_topics_threshold",
)
settings["BanTopics"] = {"topics": st_bt_topics, "threshold": st_bt_threshold}
if "Bias" in st_enabled_scanners:
st_bias_expander = st.sidebar.expander(
"Bias",
expanded=False,
)
with st_bias_expander:
st_bias_threshold = st.slider(
label="Threshold",
value=0.75,
min_value=0.0,
max_value=1.0,
step=0.05,
key="bias_threshold",
)
st_bias_match_type = st.selectbox(
"Match type", [e.value for e in BiasMatchType], index=1, key="bias_match_type"
)
settings["Bias"] = {
"threshold": st_bias_threshold,
"match_type": BiasMatchType(st_bias_match_type),
}
if "Code" in st_enabled_scanners:
st_cd_expander = st.sidebar.expander(
"Code",
expanded=False,
)
with st_cd_expander:
st_cd_languages = st.multiselect(
"Programming languages",
options=SUPPORTED_CODE_LANGUAGES,
default=["Python"],
)
st_cd_is_blocked = st.checkbox("Is blocked", value=False, key="cd_is_blocked")
settings["Code"] = {
"languages": st_cd_languages,
"is_blocked": st_cd_is_blocked,
}
if "Deanonymize" in st_enabled_scanners:
st_de_expander = st.sidebar.expander(
"Deanonymize",
expanded=False,
)
with st_de_expander:
st_de_matching_strategy = st.selectbox(
"Matching strategy", [e.value for e in DeanonymizeMatchingStrategy], index=0
)
settings["Deanonymize"] = {
"matching_strategy": DeanonymizeMatchingStrategy(st_de_matching_strategy)
}
if "JSON" in st_enabled_scanners:
st_json_expander = st.sidebar.expander(
"JSON",
expanded=False,
)
with st_json_expander:
st_json_required_elements = st.slider(
label="Required elements",
value=0,
min_value=0,
max_value=10,
step=1,
key="json_required_elements",
help="The minimum number of JSON elements that should be present",
)
st_json_repair = st.checkbox(
"Repair", value=False, help="Attempt to repair the JSON", key="json_repair"
)
settings["JSON"] = {
"required_elements": st_json_required_elements,
"repair": st_json_repair,
}
if "Language" in st_enabled_scanners:
st_lan_expander = st.sidebar.expander(
"Language",
expanded=False,
)
with st_lan_expander:
st_lan_valid_language = st.multiselect(
"Languages",
[
"ar",
"bg",
"de",
"el",
"en",
"es",
"fr",
"hi",
"it",
"ja",
"nl",
"pl",
"pt",
"ru",
"sw",
"th",
"tr",
"ur",
"vi",
"zh",
],
default=["en"],
)
st_lan_match_type = st.selectbox(
"Match type", [e.value for e in LanguageMatchType], index=1, key="lan_match_type"
)
settings["Language"] = {
"valid_languages": st_lan_valid_language,
"match_type": LanguageMatchType(st_lan_match_type),
}
if "MaliciousURLs" in st_enabled_scanners:
st_murls_expander = st.sidebar.expander(
"Malicious URLs",
expanded=False,
)
with st_murls_expander:
st_murls_threshold = st.slider(
label="Threshold",
value=0.75,
min_value=0.0,
max_value=1.0,
step=0.05,
key="murls_threshold",
)
settings["MaliciousURLs"] = {"threshold": st_murls_threshold}
if "NoRefusal" in st_enabled_scanners:
st_no_ref_expander = st.sidebar.expander(
"No refusal",
expanded=False,
)
with st_no_ref_expander:
st_no_ref_threshold = st.slider(
label="Threshold",
value=0.5,
min_value=0.0,
max_value=1.0,
step=0.05,
key="no_ref_threshold",
)
settings["NoRefusal"] = {"threshold": st_no_ref_threshold}
if "NoRefusalLight" in st_enabled_scanners:
settings["NoRefusalLight"] = {}
if "ReadingTime" in st_enabled_scanners:
st_rt_expander = st.sidebar.expander(
"Reading Time",
expanded=False,
)
with st_rt_expander:
st_rt_max_reading_time = st.slider(
label="Max reading time (in minutes)",
value=5,
min_value=0,
max_value=3600,
step=5,
key="rt_max_reading_time",
)
st_rt_truncate = st.checkbox(
"Truncate",
value=False,
help="Truncate the text to the max reading time",
key="rt_truncate",
)
settings["ReadingTime"] = {"max_time": st_rt_max_reading_time, "truncate": st_rt_truncate}
if "FactualConsistency" in st_enabled_scanners:
st_fc_expander = st.sidebar.expander(
"FactualConsistency",
expanded=False,
)
with st_fc_expander:
st_fc_minimum_score = st.slider(
label="Minimum score",
value=0.5,
min_value=0.0,
max_value=1.0,
step=0.05,
key="fc_threshold",
)
settings["FactualConsistency"] = {"minimum_score": st_fc_minimum_score}
if "Regex" in st_enabled_scanners:
st_regex_expander = st.sidebar.expander(
"Regex",
expanded=False,
)
with st_regex_expander:
st_regex_patterns = st.text_area(
"Enter patterns to ban (one per line)",
value="Bearer [A-Za-z0-9-._~+/]+",
height=200,
).split("\n")
st_regex_is_blocked = st.checkbox("Is blocked", value=True, key="regex_is_blocked")
st_regex_redact = st.checkbox(
"Redact",
value=False,
help="Replace the matched bad patterns with [REDACTED]",
key="regex_redact",
)
settings["Regex"] = {
"patterns": st_regex_patterns,
"is_blocked": st_regex_is_blocked,
"redact": st_regex_redact,
}
if "Relevance" in st_enabled_scanners:
st_rele_expander = st.sidebar.expander(
"Relevance",
expanded=False,
)
with st_rele_expander:
st_rele_threshold = st.slider(
label="Threshold",
value=0.5,
min_value=0.0,
max_value=1.0,
step=0.05,
key="rele_threshold",
)
settings["Relevance"] = {"threshold": st_rele_threshold}
if "Sensitive" in st_enabled_scanners:
st_sens_expander = st.sidebar.expander(
"Sensitive",
expanded=False,
)
with st_sens_expander:
st_sens_entity_types = st_tags(
label="Sensitive entities",
text="Type and press enter",
value=default_entity_types,
suggestions=default_entity_types
+ ["DATE_TIME", "NRP", "LOCATION", "MEDICAL_LICENSE", "US_PASSPORT"],
maxtags=30,
key="sensitive_entity_types",
)
st.caption(
"Check all supported entities: https://llm-guard.com/input_scanners/anonymize/"
)
st_sens_redact = st.checkbox("Redact", value=False, key="sens_redact")
st_sens_threshold = st.slider(
label="Threshold",
value=0.0,
min_value=0.0,
max_value=1.0,
step=0.1,
key="sens_threshold",
)
settings["Sensitive"] = {
"entity_types": st_sens_entity_types,
"redact": st_sens_redact,
"threshold": st_sens_threshold,
}
if "Sentiment" in st_enabled_scanners:
st_sent_expander = st.sidebar.expander(
"Sentiment",
expanded=False,
)
with st_sent_expander:
st_sent_threshold = st.slider(
label="Threshold",
value=-0.5,
min_value=-1.0,
max_value=1.0,
step=0.1,
key="sentiment_threshold",
help="Negative values are negative sentiment, positive values are positive sentiment",
)
settings["Sentiment"] = {"threshold": st_sent_threshold}
if "Toxicity" in st_enabled_scanners:
st_tox_expander = st.sidebar.expander(
"Toxicity",
expanded=False,
)
with st_tox_expander:
st_tox_threshold = st.slider(
label="Threshold",
value=0.5,
min_value=0.0,
max_value=1.0,
step=0.05,
key="toxicity_threshold",
)
st_tox_match_type = st.selectbox(
"Match type",
[e.value for e in ToxicityMatchType],
index=1,
key="toxicity_match_type",
)
settings["Toxicity"] = {
"threshold": st_tox_threshold,
"match_type": ToxicityMatchType(st_tox_match_type),
}
if "URLReachability" in st_enabled_scanners:
st_url_expander = st.sidebar.expander(
"URL Reachability",
expanded=False,
)
if st_url_expander:
settings["URLReachability"] = {}
if "Gibberish" in st_enabled_scanners:
st_gib_expander = st.sidebar.expander(
"Gibberish",
expanded=False,
)
with st_gib_expander:
st_gib_threshold = st.slider(
label="Threshold",
value=0.7,
min_value=0.0,
max_value=1.0,
step=0.1,
key="gib_threshold",
)
st_gib_match_type = st.selectbox(
"Match type", [e.value for e in GibberishMatchType], index=1, key="gib_match_type"
)
settings["Gibberish"] = {"match_type": st_gib_match_type, "threshold": st_gib_threshold}
return st_enabled_scanners, settings
def get_scanner(scanner_name: str, vault: Vault, settings: Dict):
logger.debug(f"Initializing {scanner_name} scanner")
if scanner_name == "Deanonymize":
settings["vault"] = vault
if scanner_name in [
"BanCode",
"BanTopics",
"Bias",
"Code",
"Gibberish",
"Language",
"LanguageSame",
"MaliciousURLs",
"NoRefusal",
"FactualConsistency",
"Relevance",
"Sensitive",
"Toxicity",
]:
settings["use_onnx"] = True
return get_scanner_by_name(scanner_name, settings)
def scan(
vault: Vault,
enabled_scanners: List[str],
settings: Dict,
prompt: str,
text: str,
fail_fast: bool = False,
) -> (str, List[Dict[str, any]]):
sanitized_output = text
results = []
status_text = "Scanning prompt..."
if fail_fast:
status_text = "Scanning prompt (fail fast mode)..."
with st.status(status_text, expanded=True) as status:
for scanner_name in enabled_scanners:
st.write(f"{scanner_name} scanner...")
scanner = get_scanner(
scanner_name, vault, settings[scanner_name] if scanner_name in settings else {}
)
start_time = time.monotonic()
sanitized_output, is_valid, risk_score = scanner.scan(prompt, sanitized_output)
end_time = time.monotonic()
results.append(
{
"scanner": scanner_name,
"is_valid": is_valid,
"risk_score": risk_score,
"took_sec": round(timedelta(seconds=end_time - start_time).total_seconds(), 2),
}
)
if fail_fast and not is_valid:
break
status.update(label="Scanning complete", state="complete", expanded=False)
return sanitized_output, results
|