forked from lancedb/lancedb
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathinstructor.py
More file actions
150 lines (127 loc) · 5.93 KB
/
instructor.py
File metadata and controls
150 lines (127 loc) · 5.93 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
# Copyright (c) 2023. LanceDB Developers
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from typing import List
import numpy as np
from ..util import attempt_import_or_raise
from .base import TextEmbeddingFunction
from .registry import register
from .utils import TEXT, weak_lru
@register("instructor")
class InstructorEmbeddingFunction(TextEmbeddingFunction):
"""
An embedding function that uses the InstructorEmbedding library. Instructor models
support multi-task learning, and can be used for a variety of tasks, including
text classification, sentence similarity, and document retrieval. If you want to
calculate customized embeddings for specific sentences, you may follow the unified
template to write instructions:
"Represent the `domain` `text_type` for `task_objective`":
* domain is optional, and it specifies the domain of the text, e.g., science,
finance, medicine, etc.
* text_type is required, and it specifies the encoding unit, e.g., sentence,
document, paragraph, etc.
* task_objective is optional, and it specifies the objective of embedding,
e.g., retrieve a document, classify the sentence, etc.
For example, if you want to calculate embeddings for a document, you may write the
instruction as follows:
"Represent the document for retrieval"
Parameters
----------
name: str
The name of the model to use. Available models are listed at
https://github.com/xlang-ai/instructor-embedding#model-list;
The default model is hkunlp/instructor-base
batch_size: int, default 32
The batch size to use when generating embeddings
device: str, default "cpu"
The device to use when generating embeddings
show_progress_bar: bool, default True
Whether to show a progress bar when generating embeddings
normalize_embeddings: bool, default True
Whether to normalize the embeddings
quantize: bool, default False
Whether to quantize the model
source_instruction: str, default "represent the document for retrieval"
The instruction for the source column
query_instruction: str, default "represent the document for retrieving the most
similar documents"
The instruction for the query
Examples
--------
import lancedb
from lancedb.pydantic import LanceModel, Vector
from lancedb.embeddings import get_registry, InstuctorEmbeddingFunction
instructor = get_registry().get("instructor").create(
source_instruction="represent the document for retrieval",
query_instruction="represent the document for retrieving the most "
"similar documents"
)
class Schema(LanceModel):
vector: Vector(instructor.ndims()) = instructor.VectorField()
text: str = instructor.SourceField()
db = lancedb.connect("~/.lancedb")
tbl = db.create_table("test", schema=Schema, mode="overwrite")
texts = [{"text": "Capitalism has been dominant in the Western world since the "
"end of feudalism, but most feel[who?] that..."},
{"text": "The disparate impact theory is especially controversial under "
"the Fair Housing Act because the Act..."},
{"text": "Disparate impact in United States labor law refers to practices "
"in employment, housing, and other areas that.."}]
tbl.add(texts)
"""
name: str = "hkunlp/instructor-base"
batch_size: int = 32
device: str = "cpu"
show_progress_bar: bool = True
normalize_embeddings: bool = True
quantize: bool = False
# convert_to_numpy: bool = True # Hardcoding this as numpy can be ingested directly
source_instruction: str = "represent the document for retrieval"
query_instruction: str = (
"represent the document for retrieving the most similar documents"
)
@weak_lru(maxsize=1)
def ndims(self):
model = self.get_model()
return model.encode("foo").shape[0]
def compute_query_embeddings(self, query: str, *args, **kwargs) -> List[np.array]:
return self.generate_embeddings([[self.query_instruction, query]])
def compute_source_embeddings(self, texts: TEXT, *args, **kwargs) -> List[np.array]:
texts = self.sanitize_input(texts)
texts_formatted = [[self.source_instruction, text] for text in texts]
return self.generate_embeddings(texts_formatted)
def generate_embeddings(self, texts: List) -> List:
model = self.get_model()
res = model.encode(
texts,
batch_size=self.batch_size,
show_progress_bar=self.show_progress_bar,
normalize_embeddings=self.normalize_embeddings,
device=self.device,
).tolist()
return res
@weak_lru(maxsize=1)
def get_model(self):
instructor_embedding = attempt_import_or_raise(
"InstructorEmbedding", "InstructorEmbedding"
)
torch = attempt_import_or_raise("torch", "torch")
model = instructor_embedding.INSTRUCTOR(self.name)
if self.quantize:
if (
"qnnpack" in torch.backends.quantized.supported_engines
): # fix for https://github.com/pytorch/pytorch/issues/29327
torch.backends.quantized.engine = "qnnpack"
model = torch.quantization.quantize_dynamic(
model, {torch.nn.Linear}, dtype=torch.qint8
)
return model