Non-autoregressive System 1 decision engine. Typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request.
Python calibrationclassificationdecision-modelhuggingfacejevmodernbertmultilingualnlppythonpytorchroutingtyped-decisionszero-shot
docs: the README's Router link points at a heading that does not exist (#236)
PerryLink8d86b5805a4a
diff --git a/README.md b/README.mdindex a11bdaec..c17dcc9d 100644--- a/README.md+++ b/README.md@@ -536,7 +536,7 @@ All of the capability on this benchmark comes from fine-tuning.Across all 51 languages the English checkpoint macro-averages **0.227** with macro ECE**0.733**, and only 23 of 51 languages clear 3x random. Khmer scores **0.000 at 95.2%-confidence**. This is why [`Router`](#model-routing-three-checkpoints-one-call) exists: the+confidence**. This is why [`Router`](#quickstart-route-mode-recommended) exists: themodel's own confidence gives no warning, so the routing decision has to be made before theforward pass.diff --git a/tests/test_packaging.py b/tests/test_packaging.pyindex 1990f3ad..9854b075 100644--- a/tests/test_packaging.py+++ b/tests/test_packaging.py@@ -93,6 +93,59 @@ missing_from_ci = []check("ci/tests every advertised Python version", missing_from_ci, [])++# --------------------------------------------------------------- markdown links+# Nothing checked these, and the README ships to PyPI and to the model card. Only+# targets inside the repository are checked: external URLs would make the suite depend+# on the network, which it must not. The README's Router link pointed at a heading that+# had been renamed, so it silently went nowhere for as long as the rename was in.+def _slug(heading):+ """GitHub's heading anchor: drop anything that is not word/space/hyphen, lowercase,+ then spaces to hyphens."""+ text = re.sub(r"[^\w\s-]", "", heading.strip().lower(), flags=re.UNICODE)+ return re.sub(r"\s+", "-", text)+++def _headings(path):+ found = set()+ for line in read(path).splitlines():+ match = re.match(r"^#{1,6}\s+(.*?)\s*$", line)+ if match:+ found.add(_slug(match.group(1)))+ return found+++_md = []+for _dirpath, _dirnames, _filenames in os.walk("."):+ # `.pytest_cache` ships a README of its own and is not part of the repository.+ _dirnames[:] = [d for d in _dirnames+ if d not in (".git", "__pycache__", "node_modules", ".pytest_cache")]+ _md.extend(os.path.normpath(os.path.join(_dirpath, f))+ for f in _filenames if f.endswith(".md"))+_md = sorted(_md)+_heading_cache = {p: _headings(p) for p in _md}++check_true("md/at least the README, BENCHMARKS and docs are scanned", len(_md) >= 8, _md)+check_true("md/no build directory scanned",+ not any(".pytest_cache" in p for p in _md), _md)++_broken_files, _broken_anchors = [], []+for _path in _md:+ for _label, _target in re.findall(r"\[([^\]]*)\]\(([^)\s]+?)(?:\s+\"[^\"]*\")?\)",+ read(_path)):+ if _target.startswith(("http://", "https://", "mailto:", "data:")):+ continue+ _tpath, _, _frag = _target.partition("#")+ _dest = os.path.normpath(os.path.join(os.path.dirname(_path), _tpath)) if _tpath else _path+ if _tpath and not os.path.exists(_dest):+ _broken_files.append("%s: [%s](%s)" % (_path, _label[:30], _target))+ continue+ if _frag and _dest.endswith(".md") and _frag not in _heading_cache.get(_dest, set()):+ _broken_anchors.append("%s: [%s](%s)" % (_path, _label[:30], _target))++check("md/no link to a file that does not exist", _broken_files, [])+check("md/no anchor that matches no heading", _broken_anchors, [])+print("\n%d passed, %d failed" % (len(PASS), len(FAIL)))for f in FAIL:print(" FAIL", f)