Spaces:
Paused
Paused
File size: 3,258 Bytes
656439d | 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 | """Contextual Retrieval — chunk-situating context prepended before embedding.
Offline: a fake LLM supplies deterministic contexts; asserts context is applied
to the embedded text but never to display/citation, and that it's gated off by
default and idempotent.
"""
from __future__ import annotations
from auralynq.config import get_settings, reload_settings
from auralynq.ingest.models import Chunk, Document, SourceType
class _FakeLLM:
name = "fake"
def __init__(self):
self.calls = 0
def generate(self, prompt, **kw):
self.calls += 1
return "This chunk is from the FRAND licensing section of the Ericsson report."
def _doc(n_chunks=3):
chunks = [
Chunk(
id=Chunk.make_id("d1", i),
doc_id="d1",
text=f"Sentence number {i} about Ericsson patents.",
ordinal=i,
source="e.pdf",
source_type=SourceType.html,
)
for i in range(n_chunks)
]
return Document(
id="d1",
source="e.pdf",
source_type=SourceType.html,
title="Ericsson",
content_hash="h",
chunks=chunks,
)
def test_embed_text_prepends_context_only_when_present():
c = Chunk(id="x", doc_id="d", text="raw body")
assert c.embed_text() == "raw body"
c.context = "situating context"
assert c.embed_text() == "situating context\n\nraw body"
# display/citation text is untouched
assert c.text == "raw body"
def test_contextual_disabled_by_default():
reload_settings()
assert get_settings().retrieval.contextual_enabled is False
def test_contextualize_sets_context_and_is_idempotent():
from auralynq.ingest.contextualize import contextualize_document
doc = _doc(3)
llm = _FakeLLM()
done = contextualize_document(doc, llm=llm)
assert done == 3
assert all(c.context for c in doc.chunks)
assert all("FRAND licensing section" in c.context for c in doc.chunks)
# embedded text carries context; raw text unchanged
assert doc.chunks[0].embed_text().startswith("This chunk is from")
assert doc.chunks[0].text == "Sentence number 0 about Ericsson patents."
# idempotent: already-set contexts are not re-generated
again = contextualize_document(doc, llm=llm)
assert again == 0 and llm.calls == 3
def test_index_documents_contextualizes_when_enabled(monkeypatch):
from auralynq.vectorstore.factory import get_store
monkeypatch.setenv("AURALYNQ_RETRIEVAL__CONTEXTUAL_ENABLED", "true")
reload_settings()
get_store.cache_clear()
llm = _FakeLLM()
monkeypatch.setattr("auralynq.llm.factory.get_llm", lambda: llm)
from auralynq.pipeline import index_documents
doc = _doc(2)
stats = index_documents([doc])
assert stats["chunks_indexed"] == 2
assert llm.calls == 2 # one per chunk
# the in-memory chunks now carry context, but display text is unchanged
assert all(c.context for c in doc.chunks)
assert doc.chunks[0].text.startswith("Sentence number 0")
# stored chunks keep the original display text (context is separate)
stored = {c.id: c for c in get_store().all_chunks()}
assert stored[doc.chunks[0].id].text == "Sentence number 0 about Ericsson patents."
|