Spaces:
Runtime error
Runtime error
File size: 8,309 Bytes
aacc29a | 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 | from __future__ import annotations
import json
import os
import sys
import tempfile
import unittest
from pathlib import Path
from unittest.mock import patch
sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
from carepath.config import Settings
from carepath.services.llm import LLMError, build_llm
from carepath.services.retrieval import RetrievedTerm
from carepath.services.scribe_local import LocalScribeLLM
def _bundle(
root: Path, *, scope: str = "research_only", correction_mode: str = "adapter"
) -> Path:
adapter_names = ("gec", "soap") if correction_mode == "adapter" else ("soap",)
for name in adapter_names:
(root / "adapters" / name).mkdir(parents=True)
manifest = {
"schema": "carepath.scribe.bundle/1",
"usage_scope": scope,
"promotion_status": "blocked_research_only",
"base_model": "Qwen/Qwen3-4B-Instruct-2507",
"adapters": {name: f"adapters/{name}" for name in adapter_names},
"correction_mode": correction_mode,
}
(root / "scribe_manifest.json").write_text(json.dumps(manifest), encoding="utf-8")
return root
class LocalScribeTests(unittest.TestCase):
def test_dual_adapter_flow_is_grounded_and_identifiable(self) -> None:
with tempfile.TemporaryDirectory() as temp:
calls: list[str] = []
def generate(adapter: str, prompt: str) -> str:
calls.append(adapter)
task = json.loads(prompt)["task"]
if task == "correct_asr_transcript":
return "Bệnh nhân dùng metformin 500 mg"
if task == "extract_grounded_clinical_facts":
transcript = "Bệnh nhân dùng metformin 500 mg"
span = "dùng metformin 500 mg"
start = transcript.index(span)
return json.dumps(
{
"facts": [
{
"type": "medication",
"value": "metformin 500 mg",
"negated": False,
"uncertain": False,
"source_span": {
"start": start,
"end": start + len(span),
"text": span,
},
}
]
}
)
return json.dumps(
{
"subjective": "metformin 500 mg",
"objective": "Chưa có thông tin khách quan.",
"assessment": "Chưa có đánh giá trong bản ghi.",
"plan": "metformin 500 mg",
"missing_information": ["Đánh giá"],
"review_required": False,
}
)
llm = LocalScribeLLM(_bundle(Path(temp)), generate_fn=generate)
terms = [RetrievedTerm("metformin", 1.0, "drug", "test")]
correction = llm.correct_transcript("Benh nhan dung metformin 500 mg", terms)
result = llm.generate_soap(correction.corrected_text, terms)
self.assertEqual(correction.provider, "scribe_local")
self.assertEqual(result.provider, "scribe_local")
self.assertTrue(result.soap.review_required)
self.assertEqual(calls, ["gec", "soap", "soap"])
ready, details = llm.readiness()
self.assertTrue(ready)
self.assertEqual(details["promotion_status"], "blocked_research_only")
self.assertEqual(details["fallback"], "disabled")
def test_unsupported_fact_and_number_fail_closed(self) -> None:
with tempfile.TemporaryDirectory() as temp:
outputs = iter(
[
json.dumps(
{
"facts": [
{
"type": "medication",
"value": "warfarin 5 mg",
"source_span": {
"start": 0,
"end": 13,
"text": "warfarin 5 mg",
},
}
]
}
)
]
)
llm = LocalScribeLLM(
_bundle(Path(temp)), generate_fn=lambda adapter, prompt: next(outputs)
)
with self.assertRaises(LLMError):
llm.generate_soap("Bệnh nhân dùng metformin 500 mg", [])
def test_writer_cannot_append_text_outside_grounded_fact_values(self) -> None:
with tempfile.TemporaryDirectory() as temp:
transcript = "Bác sĩ đánh giá viêm họng"
span = "viêm họng"
start = transcript.index(span)
outputs = iter(
[
json.dumps(
{
"facts": [
{
"type": "assessment",
"value": span,
"negated": False,
"uncertain": False,
"source_span": {
"start": start,
"end": start + len(span),
"text": span,
},
}
]
}
),
json.dumps(
{
"subjective": "Chưa có thông tin chủ quan.",
"objective": "Chưa có thông tin khách quan.",
"assessment": "viêm họng; ung thư",
"plan": "Chưa có kế hoạch trong bản ghi.",
"missing_information": [],
"review_required": True,
}
),
]
)
llm = LocalScribeLLM(
_bundle(Path(temp), correction_mode="identity"),
generate_fn=lambda adapter, prompt: next(outputs),
)
with self.assertRaisesRegex(LLMError, "outside grounded fact values"):
llm.generate_soap(transcript, [])
def test_manifest_cannot_claim_promotable_scope(self) -> None:
with tempfile.TemporaryDirectory() as temp:
with self.assertRaisesRegex(ValueError, "research_only"):
LocalScribeLLM(_bundle(Path(temp), scope="production"))
def test_soap_only_bundle_declares_identity_correction(self) -> None:
with tempfile.TemporaryDirectory() as temp:
llm = LocalScribeLLM(_bundle(Path(temp), correction_mode="identity"))
correction = llm.correct_transcript("Giữ nguyên bản ghi", [])
ready, details = llm.readiness()
self.assertTrue(ready)
self.assertEqual(correction.corrected_text, "Giữ nguyên bản ghi")
self.assertEqual(correction.provider, "scribe_local_identity")
self.assertEqual(details["adapters"], ["soap"])
def test_build_llm_requires_explicit_staging_bundle_without_fallback(self) -> None:
with tempfile.TemporaryDirectory() as temp:
_bundle(Path(temp))
with patch.dict(
os.environ,
{
"LLM_PROVIDER": "scribe_local",
"SCRIBE_BUNDLE_PATH": temp,
"LLM_FALLBACK_OFFLINE": "false",
},
clear=True,
):
llm = build_llm(Settings.from_env())
self.assertIsInstance(llm, LocalScribeLLM)
if __name__ == "__main__":
unittest.main()
|