Spaces:
Running
Running
File size: 28,407 Bytes
b210edb dfcbcc5 b210edb 58ea361 c93a7ed 58ea361 ce148d0 58ea361 c93a7ed 58ea361 c93a7ed 58ea361 c93a7ed 58ea361 2e3661e 58ea361 2e3661e 58ea361 c93a7ed 58ea361 b210edb 2f93183 dfcbcc5 2f93183 dfcbcc5 b210edb dfcbcc5 b210edb dfcbcc5 b210edb dfcbcc5 b210edb dfcbcc5 a55c5d2 dfcbcc5 2f93183 dfcbcc5 b210edb dfcbcc5 b210edb dfcbcc5 b210edb 6b60a0a dfcbcc5 a55c5d2 b210edb 6b60a0a 2f93183 6b60a0a a55c5d2 dfbcee9 58ea361 dfbcee9 a55c5d2 dfbcee9 a55c5d2 dfbcee9 2e3661e dfbcee9 b210edb dfbcee9 b210edb dfbcee9 b210edb 1af9392 b210edb 2e3661e b210edb dfbcee9 b210edb 6b60a0a b210edb dfbcee9 b210edb 340004c b210edb bbd83fb dfbcee9 b210edb bbd83fb b210edb dfbcee9 b210edb dfbcee9 b210edb dfbcee9 b210edb c88a475 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 | #!/usr/bin/env python3
"""
modeldna β HuggingFace Space
Interactive model provenance scanner.
Replaces the stale RadicalNotionAI/modelatlas-dashboard Space.
Deployed at: https://huggingface.co/spaces/RadicalNotionAI/modeldna
Custom domain: modeldna.ai (via HF Space custom domain setting)
"""
import gradio as gr
import json
import sys
import time
from pathlib import Path
# scan.py is in the same directory as app.py in both local hf_space/ and on HF
sys.path.insert(0, str(Path(__file__).parent))
from scan import scan, inspect_gguf, KNOWN_BASES, reference_count
# ββ DNA-017: scan telemetry β HF dataset (fuels derivative discovery + demand-driven ingestion)
# Guarded on LOG_TOKEN (a fine-grained WRITE token for RadicalNotionAI/modeldna-scan-log).
# No-op if the secret isn't set, so deploying this never breaks the Space.
import os as _os, json as _json, datetime as _dt
from pathlib import Path as _Path
_SCAN_LOGGER = None
_LOG_FILE = None
try:
_log_token = _os.environ.get("LOG_TOKEN")
if _log_token:
from huggingface_hub import CommitScheduler
_LOG_DIR = _Path("scan_logs"); _LOG_DIR.mkdir(exist_ok=True)
# DNA-017 fix: unique file per Space session. Space storage is ephemeral, so a single
# shared scans.jsonl got wiped on every restart and the CommitScheduler then overwrote
# the dataset copy with only post-restart rows (silent truncation). Per-session filenames
# let each run add its own file; CommitScheduler's upload_folder never deletes the others.
_session = _dt.datetime.now(_dt.timezone.utc).strftime("%Y%m%dT%H%M%SZ")
_LOG_FILE = _LOG_DIR / f"scans-{_session}-{_os.getpid()}.jsonl"
_SCAN_LOGGER = CommitScheduler(
repo_id="RadicalNotionAI/modeldna-scan-log", repo_type="dataset",
folder_path=str(_LOG_DIR), every=1, token=_log_token,
path_in_repo="data", squash_history=False,
)
print("[DNA-017] scan logging ENABLED -> RadicalNotionAI/modeldna-scan-log (flush every 1 min)", flush=True)
else:
print("[DNA-017] scan logging DISABLED: LOG_TOKEN env var not set/empty", flush=True)
except Exception as _e:
_SCAN_LOGGER = None
print(f"[DNA-017] scan logging DISABLED: scheduler init failed: {_e!r}", flush=True)
def log_scan(model_id: str, result: dict, source: str = "ui") -> None:
if _SCAN_LOGGER is None:
return
try:
v = result.get("verdict", {})
rec = {
"model_id": model_id,
"source": source, # "ui" (web) vs "internal-test" (our gradio_client) β for usage filtering
"ts": _dt.datetime.now(_dt.timezone.utc).isoformat(),
"format": result.get("format", "checkpoint"),
"architecture": v.get("architecture", "") or result.get("detected", "")[:80],
"confidence": v.get("confidence"),
"risk_score": v.get("risk_score"),
"risk_band": v.get("risk_band"),
"flag_types": [f.get("type") for f in v.get("flags", [])],
"resolved_source": result.get("resolved_source"),
"error": result.get("error"),
}
with _SCAN_LOGGER.lock:
with open(_LOG_FILE, "a") as fh:
fh.write(_json.dumps(rec) + "\n")
print(f"[DNA-017] logged scan: {model_id} ({rec['confidence']}/{rec['risk_band']})", flush=True)
except Exception as _e:
print(f"[DNA-017] log_scan failed for {model_id}: {_e!r}", flush=True)
# ββ Discovery: find derivatives that may not attribute properly ββββββββββββ
def find_unattributed_derivatives(base_match: str, scanned_id: str) -> list[dict]:
"""
Query the scan results database for models sharing the same base
that don't declare attribution to their source.
Returns models that appear derivative but lack proper attribution.
"""
try:
import psycopg2
conn = psycopg2.connect(
"postgresql:///modelatlas?host=/var/run/postgresql&port=5433&user=tim"
)
cur = conn.cursor()
# Find models in the scan results that match this base but lack attribution
# (placeholder query β will be populated as scans accumulate)
cur.execute("""
SELECT model_id, confirmed_base, has_attribution, downloads
FROM modeldna_scans
WHERE confirmed_base = %s
AND model_id != %s
AND (has_attribution = false OR has_attribution IS NULL)
ORDER BY downloads DESC NULLS LAST
LIMIT 5
""", (base_match, scanned_id))
rows = cur.fetchall()
cur.close(); conn.close()
return [{"model_id": r[0], "confirmed_base": r[1], "downloads": r[3]} for r in rows]
except Exception:
return []
def store_scan_result(result: dict) -> None:
"""Store a scan result for future derivative discovery."""
try:
import psycopg2
conn = psycopg2.connect(
"postgresql:///modelatlas?host=/var/run/postgresql&port=5433&user=tim"
)
cur = conn.cursor()
cur.execute("""
CREATE TABLE IF NOT EXISTS modeldna_scans (
id SERIAL PRIMARY KEY,
model_id TEXT UNIQUE,
confirmed_base TEXT,
confidence TEXT,
has_attribution BOOLEAN,
flag_count INT,
downloads INT,
scanned_at TIMESTAMPTZ DEFAULT now()
)
""")
v = result.get("verdict", {})
m = result.get("metadata", {})
e = result.get("evidence", {})
has_attr = bool(e.get("claimed_base"))
cur.execute("""
INSERT INTO modeldna_scans
(model_id, confirmed_base, confidence, has_attribution, flag_count, downloads)
VALUES (%s, %s, %s, %s, %s, %s)
ON CONFLICT (model_id) DO UPDATE
SET confidence=EXCLUDED.confidence,
has_attribution=EXCLUDED.has_attribution,
flag_count=EXCLUDED.flag_count,
downloads=EXCLUDED.downloads,
scanned_at=now()
""", (
result.get("model_id"),
v.get("base_model_confirmed"),
v.get("confidence"),
has_attr,
v.get("flag_count", 0),
m.get("downloads", 0),
))
conn.commit(); cur.close(); conn.close()
except Exception:
pass # graceful β don't break the scan if storage fails
_RISK_EMOJI = {"CLEAN": "π’", "LOW": "π‘", "MODERATE": "π ", "HIGH": "π΄", "UNKNOWN": "βͺ"}
_ATTRIB_FLAGS = {"IMPOSSIBLE_WEIGHT_PROVENANCE", "NAMING_MISATTRIBUTION", "NAME_MISMATCH", "UNVERIFIABLE_CLAIM"}
_DECEPTION_FLAGS = {"PROVENANCE_SCRUBBING", "IDENTITY_OVERRIDE"}
def _claimed_brand(flags: list):
"""The closed/foreign brand a deceptive name is riding on (for the headline subtitle)."""
for f in flags:
if f.get("type") in ("IMPOSSIBLE_WEIGHT_PROVENANCE", "NAMING_MISATTRIBUTION") and f.get("term"):
return f["term"]
return None
def _headline(v: dict, base_matches: list) -> tuple[str, str, str]:
"""DNA-018: verdict-first headline. Verb tracks the FINDING, emoji tracks the RISK.
Returns (emoji, verb, subtitle). Never labels a misattributed model with a green check."""
flags = v.get("flags", [])
ftypes = {f.get("type") for f in flags}
band = v.get("risk_band", "")
conf = v.get("confidence", "")
base_label = base_matches[0]["name"] if base_matches else None
brand = _claimed_brand(flags)
re = _RISK_EMOJI.get(band, "βͺ")
if ftypes & _DECEPTION_FLAGS or band == "HIGH":
if brand and base_label:
sub = f"conceals its origin β presented as {brand}, actually {base_label}"
elif base_label:
sub = f"provenance actively obscured β true base is {base_label}"
else:
sub = "provenance actively obscured"
return re, "DECEPTIVE", sub
if ftypes & _ATTRIB_FLAGS:
if brand and base_label:
sub = f'marketed as "{brand}", but the weights are {base_label}'
elif base_label:
sub = f"name overclaims β the actual base is {base_label}"
else:
sub = "the name does not match the architecture"
return re, "MISATTRIBUTED", sub
if conf == "ORIGINAL":
return "π§¬", "ORIGINAL", "first-party architecture β not a derivative of any known base"
if conf == "NEW":
return "π", "NEW", f"{v.get('base_model_confirmed','')} β not yet catalogued"
if base_label:
if band == "CLEAN":
return "π’", "VERIFIED", f"{base_label} β naming is consistent with the architecture"
return "π΅", "IDENTIFIED", base_label
return "β", "UNRECOGNIZED", "does not match any known base in the reference"
def _moat_block(v: dict, e: dict, base_matches: list) -> str:
"""DNA-018: 'How we know' β the ModelAtlas moat, stated. Independent, architecture-derived."""
n = reference_count()
n_str = f"{(n // 100) * 100:,}+" if n >= 200 else (f"{n:,}" if n else "our catalogue of")
out = ["### 𧬠How we know",
f"Independently fingerprinted against **ModelAtlas β {n_str} catalogued models**."]
if base_matches:
top = base_matches[0]
ev = "; ".join(list(dict.fromkeys(top.get("evidence", [])))[:2]) # dedup, top 2
out.append(f"**True base:** {top['name']} β {top.get('confidence','')} match ({ev}).")
out.append("Verified from the model's own architecture β **not** the uploader's declaration.")
elif v.get("confidence") in ("ORIGINAL", "NEW"):
out.append(f"**{v.get('base_model_confirmed','')}** β no derivative signature; treated as an original architecture.")
else:
out.append("No matching base architecture was found in the reference.")
return "\n\n".join(out)
def _action_block(v: dict) -> str:
"""DNA-018: 'Why it matters / what to do' β turns the verdict into an actionable reason."""
flags = v.get("flags", [])
band = v.get("risk_band", "")
if band == "UNKNOWN":
return ("### β What this means\n\n"
"**Unverified β not a clean bill of health.** We could not match this model's "
"architecture to any known base in the reference, so we can't confirm what it is "
"or where it came from. No deception flags fired, but on an unrecognized model "
"that's *absence of evidence*, not proof it's clean. Treat provenance as unverified; "
"confirm the base, license, and training claims from the source before you rely on it.")
if not flags and band in ("CLEAN", ""):
return ("### β
What this means\n\n"
"No provenance concerns detected β architecture and naming are consistent. "
"Standard due-diligence (license, intended use) still applies.")
out = ["### β οΈ Why it matters / what to do"]
for f in flags:
t = f.get("type")
if t == "IMPOSSIBLE_WEIGHT_PROVENANCE":
brand = f.get("term", "the referenced closed model")
out.append(f"- **Impossible weight claim.** {brand} weights were never publicly released, so this model cannot contain them. At best it is trained on generated outputs (distillation) β which leaves no weight-level trace, cannot be verified, and may violate the provider's terms.")
elif t == "NAMING_MISATTRIBUTION":
out.append("- **Misattribution.** The name foregrounds a brand the model has no architectural claim to, while omitting the base it is actually built from.")
elif t == "NAME_MISMATCH":
out.append("- **Name/architecture mismatch.** The name implies a different base than the weights indicate.")
elif t == "PROVENANCE_SCRUBBING":
out.append("- **Provenance scrubbing.** The chat template denies its true lineage β a deliberate step to obscure origin.")
elif t == "IDENTITY_OVERRIDE":
out.append("- **Identity override.** The model is configured to self-report a false identity.")
else:
out.append(f"- {f.get('explanation','')}")
if band == "HIGH":
out.append("**Recommendation:** avoid for anything provenance-sensitive; treat all lineage and capability claims as unproven until independently verified.")
elif band == "MODERATE":
out.append("**Recommendation:** verify the license before use and treat capability claims as unproven β the true base above is what you are actually running.")
else:
out.append("**Recommendation:** minor β note the discrepancy and verify license terms.")
return "\n\n".join(out)
def _audit_trail(result: dict) -> str:
"""DNA-018 (DeepSeek): shareable chain-of-custody β the fingerprint the verdict is derived from."""
e = result.get("evidence", {})
cs = e.get("config_signals", {}) or {}
v = result.get("verdict", {})
top = (e.get("base_matches") or [{}])[0]
lines = ["```"]
lines.append(f"model_id : {result.get('model_id','')}")
lines.append(f"scanned_at : {result.get('scanned_at','')}")
for k, label in (("model_type", "model_type"), ("vocab_size", "vocab_size"),
("hidden_size", "hidden_size"), ("num_layers", "num_layers"),
("has_mla", "has_mla")):
if cs.get(k) is not None:
lines.append(f"{label:<13}: {cs.get(k)}")
if top:
lines.append(f"matched_base : {top.get('name')} (score {top.get('score')}, {top.get('confidence')})")
lines.append(f"risk : {v.get('risk_score')}/100 {v.get('risk_band','')}")
lines.append(f"reference : ModelAtlas ({reference_count()} models)")
lines.append("engine : ModelDNA Stage-1 (config-only, no weight download)")
lines.append("```")
return "### π Fingerprint Β· audit trail\n\n_Copy as evidence β this is exactly what the verdict is derived from._\n\n" + "\n".join(lines)
def format_verdict(result: dict) -> tuple[str, str, str]:
"""DNA-018 verdict-first layout: pane 1 = VERDICT β MOAT β ACTION; pane 2 = audit trail +
similar models + badge; pane 3 = engine panel + flag detail."""
if "error" in result:
return ("β Scan Failed", f"**Error**: {result['error']}", "")
v = result.get("verdict", {})
e = result.get("evidence", {})
m = result.get("metadata", {})
flags = v.get("flags", [])
base_matches = e.get("base_matches", [])
# ββ Pane 1: the lead. Verdict headline, risk gauge, then WHY-WE-KNOW and WHAT-TO-DO.
emoji, verb, subtitle = _headline(v, base_matches)
header = f"# {emoji} {verb}\n**{subtitle}**"
if "risk_score" in v:
band = v.get("risk_band", "")
be = _RISK_EMOJI.get(band, "βͺ")
if band == "UNKNOWN":
# don't show a reassuring 0/100 for a model we couldn't identify
header += f"\n\n### {be} Provenance: UNVERIFIED β architecture not recognized"
else:
header += f"\n\n### {be} Provenance Risk: {v['risk_score']}/100 Β· {band}"
header += f"\n\n*Scanned in {result.get('elapsed_s', '?')}s Β· Stage 1 (config-only) Β· "
header += f"π₯ {m.get('downloads',0):,} Β· π {m.get('likes',0)}*"
header += "\n\n---\n\n" + _moat_block(v, e, base_matches)
header += "\n\n---\n\n" + _action_block(v)
# ββ Pane 2: the evidence & the shareable proof.
details = _audit_trail(result)
if e.get("modelatlas_similar"):
details += "\n\n### Similar verified models (ModelAtlas reference)\n"
for s in e["modelatlas_similar"][:3]:
details += f"- `{s['model_id']}`\n"
details += "\n\n---\n\n" + build_badge(result.get("model_id", ""), v)
# ββ Pane 3: the detector engines + flag detail (backup depth).
flag_text = _fmt_engine_panel(v.get("engines", [])) + _fmt_flags(flags)
return header, details, flag_text
def build_badge(model_id: str, v: dict) -> tuple:
"""DNA-004: zero-backend shields.io badge + copyable model-card snippet."""
from urllib.parse import quote
score = v.get("risk_score"); band = v.get("risk_band", "")
color = {"CLEAN": "brightgreen", "LOW": "yellowgreen", "MODERATE": "orange",
"HIGH": "red", "UNKNOWN": "lightgrey"}.get(band, "blue")
if band == "UNKNOWN":
msg = "unverified"
elif score is None:
msg = "scanned"
elif band == "CLEAN":
msg = "clean"
else:
msg = f"{band.lower()} risk {score} of 100"
badge = f"https://img.shields.io/badge/{quote('ModelDNA')}-{quote(msg)}-{color}"
space = "https://huggingface.co/spaces/RadicalNotionAI/modeldna"
md = f"[]({space})"
block = ("### π Share / embed this result\n\n"
f"\n\n"
"Add to your model card (Markdown):\n\n"
f"```\n{md}\n```")
return block
_GENERIC_TOKENS = {"qwen", "qwen2", "qwen3", "qwen35", "llama", "gemma", "mistral",
"phi", "deepseek", "model", "base", "instruct"}
def build_lineage(model_id: str, result: dict) -> str:
"""DNA-002 v1: lineage view β where it came from + the derivative/quant family it spawned.
Uses the HF API (reachable from the Space) to find same-family models by name."""
import os, re
from huggingface_hub import HfApi
v = result.get("verdict", {})
org, short = (model_id.split("/", 1) + [""])[:2] if "/" in model_id else ("", model_id)
came_from = result.get("resolved_source") or (v.get("base_model_confirmed") or "").strip()
# distinctive search token = first name segment, unless it's a generic base word
token = re.split(r"[-_.]", short)[0]
if len(token) < 4 or token.lower() in _GENERIC_TOKENS:
token = short
api = HfApi(token=os.environ.get("HF_TOKEN"))
try:
found = list(api.list_models(search=token, limit=60))
except Exception:
found = []
groups = {"GGUF": [], "MLX": [], "AWQ/GPTQ": [], "other quant": [], "fine-tune / variant": []}
for m in found:
if m.id.lower() == model_id.lower():
continue
low = m.id.lower()
dl = getattr(m, "downloads", 0) or 0
if "gguf" in low: g = "GGUF"
elif "mlx" in low: g = "MLX"
elif "awq" in low or "gptq" in low: g = "AWQ/GPTQ"
elif any(q in low for q in ("nvfp4", "fp8", "int4", "int8", "-4bit", "-8bit", "exl2")): g = "other quant"
else: g = "fine-tune / variant"
groups[g].append((m.id, dl))
total = sum(len(x) for x in groups.values())
if total == 0 and not came_from:
return ("### π³ Lineage & Derivatives\n\nNo related models found yet. "
"As similar models are scanned, the family tree grows here.")
out = "### π³ Lineage & Derivatives\n\n"
if came_from and came_from not in ("Unknown", ""):
out += f"**β¬ Descends from:** `{came_from}`\n\n"
out += f"**β¬ Derivative family** (`{token}` β {total} related models found on HF):\n\n"
for g, items in groups.items():
if not items:
continue
items.sort(key=lambda x: -x[1])
out += f"- **{g}** ({len(items)}): "
out += ", ".join(f"`{i}` ({d:,})" for i, d in items[:4])
if len(items) > 4:
out += f", +{len(items)-4} more"
out += "\n"
out += ("\n*Downloads in parentheses. Quants inherit the source's provenance β a scrub or "
"misattribution in the parent travels to every derivative here.*")
return out
def _fmt_engine_panel(engines: list) -> str:
if not engines:
return ""
t = "### π¬ Detector Engines\n\n"
for e in engines:
t += f"- {e['status']} Β· **{e['engine']}** β {e['detail']}\n"
return t + "\n---\n\n"
def _fmt_flags(flags: list) -> str:
if not flags:
return "### β
No Flags\n\nNo suspicious claims detected in model name or metadata."
t = f"### β οΈ {len(flags)} Flag(s) Found\n\n"
for f in flags:
sev = f.get("severity", "")
t += f"**[{f['type']}]**" + (f" _(severity: {sev})_" if sev else "")
t += f"\n\n{f['explanation']}\n\n---\n\n"
return t
def format_gguf_scan(result: dict) -> tuple[str, str, str, str]:
"""Render a GGUF detection: what it is + the optional follow-up offers (as text)."""
src = result.get("resolved_source")
header = ("π¦ **GGUF quantization detected**\n\n"
"*A quant is the same weights at lower precision β its provenance equals its source.*")
details = "### What this is\n" + result.get("detected", "") + "\n\n"
if src:
details += f"**Resolved full-weights source:** `{src}`\n\n"
details += "_Use the buttons below to inspect the GGUF's embedded metadata or scan the full-weights source._"
disc = ("### π Derivative Discovery\n\nGGUF quants inherit their source's provenance "
"(and any identity scrub baked into the chat template). Inspect the metadata or "
"scan the source for the full picture.")
return header, details, "", disc
def format_inspection(result: dict) -> tuple[str, str, str, str]:
if "error" in result:
return "β GGUF inspection failed", f"**Error**: {result['error']}", "", ""
v = result.get("verdict", {}); ev = result.get("evidence", {})
header = (f"π¦ **GGUF metadata inspection** β {v.get('architecture','')}\n\n"
"*Read from the GGUF header only β no weights downloaded.*")
if "risk_score" in v:
be = {"CLEAN": "π’", "LOW": "π‘", "MODERATE": "π ", "HIGH": "π΄"}.get(v.get("risk_band",""), "βͺ")
header += f"\n\n### {be} Provenance Risk: {v['risk_score']}/100 Β· {v.get('risk_band','')}"
details = ("### GGUF header\n"
f"- **name:** {ev.get('gguf_general.name')}\n"
f"- **architecture:** `{ev.get('gguf_architecture')}`\n"
f"- **chat template present:** {ev.get('chat_template_present')}\n"
f"- **file:** `{result.get('gguf_file')}`\n")
return header, details, _fmt_engine_panel(v.get("engines", [])) + _fmt_flags(v.get("flags", [])), ""
def inspect_action(gguf_id: str):
"""Optional action (offer #1): inspect the GGUF's embedded metadata."""
return format_inspection(inspect_gguf(gguf_id))
_HIDE = None # placeholder; gr.update created inline
def run_scan(model_id: str, request: gr.Request = None):
"""Main scan. Returns 8 outputs: 4 markdown panes + GGUF action row/button updates + 2 states."""
import gradio as _gr
model_id = model_id.strip()
hide_row = _gr.update(visible=False); hide_btn = _gr.update(visible=False)
if not model_id:
return "Enter a HuggingFace model ID above.", "", "", "", hide_row, hide_btn, "", ""
if "huggingface.co/" in model_id:
model_id = model_id.split("huggingface.co/")[-1].strip("/")
# Normalize + validate: take first line/token, drop URL query/fragment, keep leading
# <org>/<name> path segments. Rejects pasted blobs/prose so telemetry stays clean (DNA-017).
model_id = model_id.splitlines()[0].strip().split()[0] if model_id.split() else ""
model_id = model_id.split("?")[0].split("#")[0].strip("/")
parts = model_id.split("/")
if len(parts) >= 2:
model_id = "/".join(parts[:2]) # spaces/datasets prefixes or trailing paths β org/name
import re as _re
if not _re.fullmatch(r"[A-Za-z0-9._-]+/[A-Za-z0-9._-]+", model_id):
msg = ("**Not a valid model ID.** Enter a HuggingFace repo as `org/name` "
"(e.g. `Qwen/Qwen3.5-27B`) or paste its huggingface.co URL.")
return msg, "", "", "", hide_row, hide_btn, "", ""
result = scan(model_id)
store_scan_result(result)
# DNA-017: source tag for usage filtering β "ui" for web visitors; our gradio_client test
# calls set the X-ModelDNA-Source header so they can be excluded from external-usage counts.
_src = "ui"
try:
if request is not None:
_src = (request.headers.get("x-modeldna-source") or "ui").lower()[:32]
except Exception:
_src = "ui"
log_scan(model_id, result, _src) # telemetry to HF dataset (no-op without LOG_TOKEN)
# GGUF: show detection + reveal the optional follow-up buttons
if str(result.get("format", "")).startswith("GGUF"):
h, d, fl, disc = format_gguf_scan(result)
src = result.get("resolved_source") or ""
src_btn = _gr.update(visible=bool(src), value=f"π¬ Scan source: {src}") if src else hide_btn
return h, d, fl, disc, _gr.update(visible=True), src_btn, src, model_id
header, details, flags = format_verdict(result)
# DNA-002: lineage & derivative family (HF-API-backed, works on the Space)
try:
discovery = build_lineage(model_id, result)
except Exception:
discovery = "### π³ Lineage & Derivatives\n\n(temporarily unavailable)"
return header, details, flags, discovery, hide_row, hide_btn, "", ""
# ββ Gradio UI ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
EXAMPLES = [
"Qwen/Qwen3.5-27B",
"Jackrong/Qwen3.5-35B-A3B-Claude-4.6-Opus-Reasoning-Distilled",
"poolside/Laguna-XS.2",
"deepseek-ai/DeepSeek-R1",
"mistralai/Mistral-Medium-3.5-128B",
]
CSS = """
.gradio-container { max-width: 900px !important; margin: 0 auto; }
.verdict-header { font-size: 1.2em; }
footer { display: none; }
"""
with gr.Blocks(
title="ModelDNA β AI Model Provenance",
theme=gr.themes.Ocean(),
css=CSS,
) as demo:
gr.Markdown("""
# 𧬠ModelDNA
### The DNA test for AI models β verify provenance before you download
*Powered by ModelAtlas Β· a RadicalNotion product*
> **Works with:** standard HuggingFace checkpoints (safetensors / PyTorch bin) **and GGUF quants**
> (header-only metadata inspection + source resolution). Not yet supported: private/gated models.
> No weight download needed β Stage 1 reads config.json (or the GGUF header) only.
---
""")
with gr.Row():
model_input = gr.Textbox(
label="HuggingFace Model ID or URL",
placeholder="e.g. Qwen/Qwen3.5-27B (not GGUF β use the original checkpoint)",
scale=4,
)
scan_btn = gr.Button("π¬ Scan", variant="primary", scale=1)
gr.Examples(
examples=EXAMPLES,
inputs=model_input,
label="Try these examples",
)
gr.Markdown("---")
with gr.Row():
header_out = gr.Markdown(label="Verdict")
with gr.Row():
with gr.Column():
details_out = gr.Markdown(label="Evidence")
with gr.Column():
flags_out = gr.Markdown(label="Flags")
# GGUF optional follow-up actions (hidden until a GGUF is scanned)
source_state = gr.State("")
gguf_id_state = gr.State("")
with gr.Row(visible=False) as gguf_actions:
inspect_btn = gr.Button("π Inspect GGUF metadata (no download)")
source_btn = gr.Button("π¬ Scan full-weights source", variant="primary")
gr.Markdown("---")
discovery_out = gr.Markdown(label="Derivative Discovery")
gr.Markdown("""
---
*Stage 1 (architecture screening): free, unlimited, no weight download needed.*
*Stage 2 (weight-level analysis): coming soon β deeper confirmation.*
*[modeldna.ai](https://modeldna.ai) Β· [RadicalNotionAI on HF](https://huggingface.co/RadicalNotionAI)*
""")
_scan_outputs = [header_out, details_out, flags_out, discovery_out,
gguf_actions, source_btn, source_state, gguf_id_state]
scan_btn.click(fn=run_scan, inputs=[model_input], outputs=_scan_outputs)
model_input.submit(fn=run_scan, inputs=[model_input], outputs=_scan_outputs)
# Optional GGUF follow-ups (user-triggered)
inspect_btn.click(
fn=inspect_action, inputs=[gguf_id_state],
outputs=[header_out, details_out, flags_out, discovery_out],
)
source_btn.click( # scan the resolved full-weights source
fn=run_scan, inputs=[source_state], outputs=_scan_outputs,
)
if __name__ == "__main__":
demo.launch()
|