bvandeusen
52d6a8ed53
Replaces the Ollama HTTP get_embedding with a fastembed.TextEmbedding singleton loaded lazily on first call. Model: BAAI/bge-small-en-v1.5 (384-dim), cached to /data/fastembed-cache. Public API unchanged: - get_embedding(text, model=None) — `model` now silently ignored - upsert_note_embedding - semantic_search_notes - backfill_note_embeddings _cosine_similarity gains a defensive length-mismatch check so any stale 768-dim row that survived the migration is treated as 0.0 similarity rather than crashing zip(). The Ollama client dep stays in pyproject for now (other services still use it); Phase 7 removes it once chat/journal/curator are gone. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
91 lines
3.2 KiB
Python
91 lines
3.2 KiB
Python
"""Tests for services.embeddings — fastembed backend.
|
|
|
|
We don't actually load the fastembed model in tests (heavy download).
|
|
Instead, mock _get_model to return a fake that produces deterministic vectors.
|
|
"""
|
|
from unittest.mock import AsyncMock, MagicMock, patch
|
|
|
|
import pytest
|
|
|
|
from fabledassistant.services.embeddings import (
|
|
_cosine_similarity, get_embedding,
|
|
)
|
|
|
|
|
|
# ─── cosine_similarity (pure logic) ──────────────────────────────────────────
|
|
|
|
|
|
def test_cosine_similarity_orthogonal_is_zero():
|
|
assert _cosine_similarity([1.0, 0.0], [0.0, 1.0]) == 0.0
|
|
|
|
|
|
def test_cosine_similarity_identical_is_one():
|
|
assert _cosine_similarity([1.0, 0.0], [1.0, 0.0]) == pytest.approx(1.0)
|
|
|
|
|
|
def test_cosine_similarity_opposite_is_negative_one():
|
|
assert _cosine_similarity([1.0, 0.0], [-1.0, 0.0]) == pytest.approx(-1.0)
|
|
|
|
|
|
def test_cosine_similarity_zero_length_safe():
|
|
"""Zero-magnitude vector must not divide-by-zero."""
|
|
assert _cosine_similarity([0.0, 0.0], [1.0, 0.0]) == 0.0
|
|
assert _cosine_similarity([1.0, 0.0], [0.0, 0.0]) == 0.0
|
|
|
|
|
|
def test_cosine_similarity_mismatched_dim_returns_zero():
|
|
"""Cross-migration safety: a 768-dim vs 384-dim comparison must not crash."""
|
|
assert _cosine_similarity([1.0] * 5, [1.0] * 3) == 0.0
|
|
|
|
|
|
def test_cosine_similarity_empty_inputs():
|
|
assert _cosine_similarity([], []) == 0.0
|
|
assert _cosine_similarity([], [1.0]) == 0.0
|
|
|
|
|
|
# ─── get_embedding (fastembed path) ──────────────────────────────────────────
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_get_embedding_returns_list_of_floats():
|
|
"""get_embedding wraps the embedder; the result is a Python list of floats."""
|
|
fake_vec = MagicMock()
|
|
fake_vec.tolist.return_value = [0.1, 0.2, 0.3, 0.4]
|
|
fake_embedder = MagicMock()
|
|
fake_embedder.embed = MagicMock(return_value=iter([fake_vec]))
|
|
with patch(
|
|
"fabledassistant.services.embeddings._get_model",
|
|
AsyncMock(return_value=fake_embedder),
|
|
):
|
|
out = await get_embedding("hello world")
|
|
assert out == [0.1, 0.2, 0.3, 0.4]
|
|
fake_embedder.embed.assert_called_once_with(["hello world"])
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_get_embedding_ignores_legacy_model_param():
|
|
"""The `model` kwarg is preserved for backward-compat but should not affect
|
|
the fastembed call."""
|
|
fake_vec = MagicMock()
|
|
fake_vec.tolist.return_value = [0.0]
|
|
fake_embedder = MagicMock()
|
|
fake_embedder.embed = MagicMock(return_value=iter([fake_vec]))
|
|
with patch(
|
|
"fabledassistant.services.embeddings._get_model",
|
|
AsyncMock(return_value=fake_embedder),
|
|
):
|
|
out = await get_embedding("x", model="ignored-model-name")
|
|
assert out == [0.0]
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_get_embedding_propagates_model_load_failures():
|
|
"""If fastembed can't initialize, the error propagates — callers catch
|
|
and degrade to keyword search."""
|
|
with patch(
|
|
"fabledassistant.services.embeddings._get_model",
|
|
AsyncMock(side_effect=RuntimeError("model load failed")),
|
|
):
|
|
with pytest.raises(RuntimeError, match="model load failed"):
|
|
await get_embedding("x")
|