forked from lancedb/lancedb
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathtest_embeddings.py
More file actions
393 lines (318 loc) · 12.1 KB
/
test_embeddings.py
File metadata and controls
393 lines (318 loc) · 12.1 KB
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
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright The LanceDB Authors
from typing import List, Union
from unittest.mock import MagicMock, patch
import lance
import lancedb
import numpy as np
import pyarrow as pa
import pytest
import pandas as pd
from lancedb.conftest import MockTextEmbeddingFunction
from lancedb.embeddings import (
EmbeddingFunctionConfig,
EmbeddingFunctionRegistry,
with_embeddings,
)
from lancedb.embeddings.base import TextEmbeddingFunction
from lancedb.embeddings.registry import get_registry, register
from lancedb.embeddings.utils import retry
from lancedb.pydantic import LanceModel, Vector
def mock_embed_func(input_data):
return [np.random.randn(128).tolist() for _ in range(len(input_data))]
def test_with_embeddings():
for wrap_api in [True, False]:
data = pa.Table.from_arrays(
[
pa.array(["foo", "bar"]),
pa.array([10.0, 20.0]),
],
names=["text", "price"],
)
data = with_embeddings(mock_embed_func, data, wrap_api=wrap_api)
assert data.num_columns == 3
assert data.num_rows == 2
assert data.column_names == ["text", "price", "vector"]
assert data.column("text").to_pylist() == ["foo", "bar"]
assert data.column("price").to_pylist() == [10.0, 20.0]
def test_embedding_function(tmp_path):
registry = EmbeddingFunctionRegistry.get_instance()
# let's create a table
table = pa.table(
{
"text": pa.array(["hello world", "goodbye world"]),
"vector": [np.random.randn(10), np.random.randn(10)],
}
)
conf = EmbeddingFunctionConfig(
source_column="text",
vector_column="vector",
function=MockTextEmbeddingFunction(),
)
metadata = registry.get_table_metadata([conf])
table = table.replace_schema_metadata(metadata)
# Write it to disk
lance.write_dataset(table, tmp_path / "test.lance")
# Load this back
ds = lance.dataset(tmp_path / "test.lance")
# can we get the serialized version back out?
configs = registry.parse_functions(ds.schema.metadata)
conf = configs["vector"]
func = conf.function
actual = func.compute_query_embeddings("hello world")
# And we make sure we can call it
expected = func.compute_query_embeddings("hello world")
assert np.allclose(actual, expected)
def test_embedding_with_bad_results(tmp_path):
@register("null-embedding")
class NullEmbeddingFunction(TextEmbeddingFunction):
def ndims(self):
return 128
def generate_embeddings(
self, texts: Union[List[str], np.ndarray]
) -> list[Union[np.array, None]]:
# Return None, which is bad if field is non-nullable
a = [
np.full(self.ndims(), np.nan)
if i % 2 == 0
else np.random.randn(self.ndims())
for i in range(len(texts))
]
return a
db = lancedb.connect(tmp_path)
registry = EmbeddingFunctionRegistry.get_instance()
model = registry.get("null-embedding").create()
class Schema(LanceModel):
text: str = model.SourceField()
vector: Vector(model.ndims()) = model.VectorField()
table = db.create_table("test", schema=Schema, mode="overwrite")
with pytest.raises(ValueError):
# Default on_bad_vectors is "error"
table.add([{"text": "hello world"}])
table.add(
[{"text": "hello world"}, {"text": "bar"}],
on_bad_vectors="drop",
)
df = table.to_pandas()
assert len(table) == 1
assert df.iloc[0]["text"] == "bar"
@register("nan-embedding")
class NanEmbeddingFunction(TextEmbeddingFunction):
def ndims(self):
return 128
def generate_embeddings(
self, texts: Union[List[str], np.ndarray]
) -> list[Union[np.array, None]]:
# Return NaN to produce bad vectors
return [
[np.NAN] * 128 if i % 2 == 0 else np.random.randn(self.ndims())
for i in range(len(texts))
]
db = lancedb.connect(tmp_path)
registry = EmbeddingFunctionRegistry.get_instance()
model = registry.get("nan-embedding").create()
table = db.create_table("test2", schema=Schema, mode="overwrite")
table.alter_columns(dict(path="vector", nullable=True))
table.add(
[{"text": "hello world"}, {"text": "bar"}],
on_bad_vectors="null",
)
assert len(table) == 2
tbl = table.to_arrow()
assert tbl["vector"].null_count == 1
def test_with_existing_vectors(tmp_path):
@register("mock-embedding")
class MockEmbeddingFunction(TextEmbeddingFunction):
def ndims(self):
return 128
def generate_embeddings(
self, texts: Union[List[str], np.ndarray]
) -> List[np.array]:
return [np.random.randn(self.ndims()).tolist() for _ in range(len(texts))]
registry = get_registry()
model = registry.get("mock-embedding").create()
class Schema(LanceModel):
text: str = model.SourceField()
vector: Vector(model.ndims()) = model.VectorField()
db = lancedb.connect(tmp_path)
tbl = db.create_table("test", schema=Schema, mode="overwrite")
tbl.add([{"text": "hello world", "vector": np.zeros(128).tolist()}])
embeddings = tbl.to_arrow()["vector"].to_pylist()
assert not np.any(embeddings), "all zeros"
def test_embedding_function_with_pandas(tmp_path):
@register("mock-embedding")
class _MockEmbeddingFunction(TextEmbeddingFunction):
def ndims(self):
return 128
def generate_embeddings(
self, texts: Union[List[str], np.ndarray]
) -> List[np.array]:
return [np.random.randn(self.ndims()).tolist() for _ in range(len(texts))]
registery = get_registry()
func = registery.get("mock-embedding").create()
class TestSchema(LanceModel):
text: str = func.SourceField()
val: int
vector: Vector(func.ndims()) = func.VectorField()
df = pd.DataFrame(
{
"text": ["hello world", "goodbye world"],
"val": [1, 2],
}
)
db = lancedb.connect(tmp_path)
tbl = db.create_table("test", schema=TestSchema, mode="overwrite", data=df)
schema = tbl.schema
assert schema.field("text").type == pa.string()
assert schema.field("val").type == pa.int64()
assert schema.field("vector").type == pa.list_(pa.float32(), 128)
df = pd.DataFrame(
{
"text": ["extra", "more"],
"val": [4, 5],
}
)
tbl.add(df)
assert tbl.count_rows() == 4
embeddings = tbl.to_arrow()["vector"]
assert embeddings.null_count == 0
df = pd.DataFrame(
{
"text": ["with", "embeddings"],
"val": [6, 7],
"vector": [np.zeros(128).tolist(), np.zeros(128).tolist()],
}
)
tbl.add(df)
embeddings = tbl.search().where("val > 5").to_arrow()["vector"].to_pylist()
assert not np.any(embeddings), "all zeros"
def test_multiple_embeddings_for_pandas(tmp_path):
@register("mock-embedding")
class MockFunc1(TextEmbeddingFunction):
def ndims(self):
return 128
def generate_embeddings(
self, texts: Union[List[str], np.ndarray]
) -> List[np.array]:
return [np.random.randn(self.ndims()).tolist() for _ in range(len(texts))]
@register("mock-embedding2")
class MockFunc2(TextEmbeddingFunction):
def ndims(self):
return 512
def generate_embeddings(
self, texts: Union[List[str], np.ndarray]
) -> List[np.array]:
return [np.random.randn(self.ndims()).tolist() for _ in range(len(texts))]
registery = get_registry()
func1 = registery.get("mock-embedding").create()
func2 = registery.get("mock-embedding2").create()
class TestSchema(LanceModel):
text: str = func1.SourceField()
val: int
vec1: Vector(func1.ndims()) = func1.VectorField()
prompt: str = func2.SourceField()
vec2: Vector(func2.ndims()) = func2.VectorField()
df = pd.DataFrame(
{
"text": ["hello world", "goodbye world"],
"val": [1, 2],
"prompt": ["hello", "goodbye"],
}
)
db = lancedb.connect(tmp_path)
tbl = db.create_table("test", schema=TestSchema, mode="overwrite", data=df)
schema = tbl.schema
assert schema.field("text").type == pa.string()
assert schema.field("val").type == pa.int64()
assert schema.field("vec1").type == pa.list_(pa.float32(), 128)
assert schema.field("prompt").type == pa.string()
assert schema.field("vec2").type == pa.list_(pa.float32(), 512)
assert tbl.count_rows() == 2
@pytest.mark.slow
def test_embedding_function_rate_limit(tmp_path):
def _get_schema_from_model(model):
class Schema(LanceModel):
text: str = model.SourceField()
vector: Vector(model.ndims()) = model.VectorField()
return Schema
db = lancedb.connect(tmp_path)
registry = EmbeddingFunctionRegistry.get_instance()
model = registry.get("test-rate-limited").create(max_retries=0)
schema = _get_schema_from_model(model)
table = db.create_table("test", schema=schema, mode="overwrite")
table.add([{"text": "hello world"}])
with pytest.raises(Exception):
table.add([{"text": "hello world"}])
assert len(table) == 1
model = registry.get("test-rate-limited").create()
schema = _get_schema_from_model(model)
table = db.create_table("test", schema=schema, mode="overwrite")
table.add([{"text": "hello world"}])
table.add([{"text": "hello world"}])
assert len(table) == 2
def test_add_optional_vector(tmp_path):
@register("mock-embedding")
class MockEmbeddingFunction(TextEmbeddingFunction):
def ndims(self):
return 128
def generate_embeddings(
self, texts: Union[List[str], np.ndarray]
) -> List[np.array]:
"""
Generate the embeddings for the given texts
"""
return [np.random.randn(self.ndims()).tolist() for _ in range(len(texts))]
registry = get_registry()
model = registry.get("mock-embedding").create()
class LanceSchema(LanceModel):
id: str
vector: Vector(model.ndims()) = model.VectorField(default=None)
text: str = model.SourceField()
db = lancedb.connect(tmp_path)
tbl = db.create_table("optional_vector", schema=LanceSchema)
# add works
expected = LanceSchema(id="id", text="text")
tbl.add([expected])
assert not (np.abs(tbl.to_pandas()["vector"][0]) < 1e-6).all()
@pytest.mark.parametrize(
"embedding_type",
[
"openai",
"sentence-transformers",
"huggingface",
"ollama",
"cohere",
"instructor",
"voyageai",
],
)
def test_embedding_function_safe_model_dump(embedding_type):
registry = get_registry()
# Note: Some embedding types might require specific parameters
try:
model = registry.get(embedding_type).create()
except Exception as e:
pytest.skip(f"Skipping {embedding_type} due to error: {str(e)}")
dumped_model = model.safe_model_dump()
assert all(
not k.startswith("_") for k in dumped_model.keys()
), f"{embedding_type}: Dumped model contains keys starting with underscore"
assert (
"max_retries" in dumped_model
), f"{embedding_type}: Essential field 'max_retries' is missing from dumped model"
assert isinstance(
dumped_model, dict
), f"{embedding_type}: Dumped model is not a dictionary"
for key in model.__dict__:
if key.startswith("_"):
assert key not in dumped_model, (
f"{embedding_type}: Private attribute '{key}' "
f"is present in dumped model"
)
@patch("time.sleep")
def test_retry(mock_sleep):
test_function = MagicMock(side_effect=[Exception] * 9 + ["result"])
test_function = retry()(test_function)
result = test_function()
assert mock_sleep.call_count == 9
assert result == "result"