haystack

1
import os
2
from typing import List
3
from haystack.utils.auth import Secret
4

5
import numpy as np
6
import pytest
7

8
from haystack import Document
9
from haystack.components.embedders.openai_document_embedder import OpenAIDocumentEmbedder
10

11

12
def mock_openai_response(input: List[str], model: str = "text-embedding-ada-002", **kwargs) -> dict:
13
    dict_response = {
14
        "object": "list",
15
        "data": [
16
            {"object": "embedding", "index": i, "embedding": np.random.rand(1536).tolist()} for i in range(len(input))
17
        ],
18
        "model": model,
19
        "usage": {"prompt_tokens": 4, "total_tokens": 4},
20
    }
21

22
    return dict_response
23

24

25
class TestOpenAIDocumentEmbedder:
26
    def test_init_default(self, monkeypatch):
27
        monkeypatch.setenv("OPENAI_API_KEY", "fake-api-key")
28
        embedder = OpenAIDocumentEmbedder()
29
        assert embedder.model == "text-embedding-ada-002"
30
        assert embedder.organization is None
31
        assert embedder.prefix == ""
32
        assert embedder.suffix == ""
33
        assert embedder.batch_size == 32
34
        assert embedder.progress_bar is True
35
        assert embedder.meta_fields_to_embed == []
36
        assert embedder.embedding_separator == "\n"
37

38
    def test_init_with_parameters(self):
39
        embedder = OpenAIDocumentEmbedder(
40
            api_key=Secret.from_token("fake-api-key"),
41
            model="model",
42
            organization="my-org",
43
            prefix="prefix",
44
            suffix="suffix",
45
            batch_size=64,
46
            progress_bar=False,
47
            meta_fields_to_embed=["test_field"],
48
            embedding_separator=" | ",
49
        )
50
        assert embedder.organization == "my-org"
51
        assert embedder.model == "model"
52
        assert embedder.prefix == "prefix"
53
        assert embedder.suffix == "suffix"
54
        assert embedder.batch_size == 64
55
        assert embedder.progress_bar is False
56
        assert embedder.meta_fields_to_embed == ["test_field"]
57
        assert embedder.embedding_separator == " | "
58

59
    def test_init_fail_wo_api_key(self, monkeypatch):
60
        monkeypatch.delenv("OPENAI_API_KEY", raising=False)
61
        with pytest.raises(ValueError, match="None of the .* environment variables are set"):
62
            OpenAIDocumentEmbedder()
63

64
    def test_to_dict(self, monkeypatch):
65
        monkeypatch.setenv("OPENAI_API_KEY", "fake-api-key")
66
        component = OpenAIDocumentEmbedder()
67
        data = component.to_dict()
68
        assert data == {
69
            "type": "haystack.components.embedders.openai_document_embedder.OpenAIDocumentEmbedder",
70
            "init_parameters": {
71
                "api_key": {"env_vars": ["OPENAI_API_KEY"], "strict": True, "type": "env_var"},
72
                "api_base_url": None,
73
                "model": "text-embedding-ada-002",
74
                "dimensions": None,
75
                "organization": None,
76
                "prefix": "",
77
                "suffix": "",
78
                "batch_size": 32,
79
                "progress_bar": True,
80
                "meta_fields_to_embed": [],
81
                "embedding_separator": "\n",
82
            },
83
        }
84

85
    def test_to_dict_with_custom_init_parameters(self, monkeypatch):
86
        monkeypatch.setenv("ENV_VAR", "fake-api-key")
87
        component = OpenAIDocumentEmbedder(
88
            api_key=Secret.from_env_var("ENV_VAR", strict=False),
89
            model="model",
90
            organization="my-org",
91
            prefix="prefix",
92
            suffix="suffix",
93
            batch_size=64,
94
            progress_bar=False,
95
            meta_fields_to_embed=["test_field"],
96
            embedding_separator=" | ",
97
        )
98
        data = component.to_dict()
99
        assert data == {
100
            "type": "haystack.components.embedders.openai_document_embedder.OpenAIDocumentEmbedder",
101
            "init_parameters": {
102
                "api_key": {"env_vars": ["ENV_VAR"], "strict": False, "type": "env_var"},
103
                "api_base_url": None,
104
                "model": "model",
105
                "dimensions": None,
106
                "organization": "my-org",
107
                "prefix": "prefix",
108
                "suffix": "suffix",
109
                "batch_size": 64,
110
                "progress_bar": False,
111
                "meta_fields_to_embed": ["test_field"],
112
                "embedding_separator": " | ",
113
            },
114
        }
115

116
    def test_prepare_texts_to_embed_w_metadata(self):
117
        documents = [
118
            Document(content=f"document number {i}:\ncontent", meta={"meta_field": f"meta_value {i}"}) for i in range(5)
119
        ]
120

121
        embedder = OpenAIDocumentEmbedder(
122
            api_key=Secret.from_token("fake-api-key"), meta_fields_to_embed=["meta_field"], embedding_separator=" | "
123
        )
124

125
        prepared_texts = embedder._prepare_texts_to_embed(documents)
126

127
        # note that newline is replaced by space
128
        assert prepared_texts == [
129
            "meta_value 0 | document number 0: content",
130
            "meta_value 1 | document number 1: content",
131
            "meta_value 2 | document number 2: content",
132
            "meta_value 3 | document number 3: content",
133
            "meta_value 4 | document number 4: content",
134
        ]
135

136
    def test_prepare_texts_to_embed_w_suffix(self):
137
        documents = [Document(content=f"document number {i}") for i in range(5)]
138

139
        embedder = OpenAIDocumentEmbedder(
140
            api_key=Secret.from_token("fake-api-key"), prefix="my_prefix ", suffix=" my_suffix"
141
        )
142

143
        prepared_texts = embedder._prepare_texts_to_embed(documents)
144

145
        assert prepared_texts == [
146
            "my_prefix document number 0 my_suffix",
147
            "my_prefix document number 1 my_suffix",
148
            "my_prefix document number 2 my_suffix",
149
            "my_prefix document number 3 my_suffix",
150
            "my_prefix document number 4 my_suffix",
151
        ]
152

153
    def test_run_wrong_input_format(self):
154
        embedder = OpenAIDocumentEmbedder(api_key=Secret.from_token("fake-api-key"))
155

156
        # wrong formats
157
        string_input = "text"
158
        list_integers_input = [1, 2, 3]
159

160
        with pytest.raises(TypeError, match="OpenAIDocumentEmbedder expects a list of Documents as input"):
161
            embedder.run(documents=string_input)
162

163
        with pytest.raises(TypeError, match="OpenAIDocumentEmbedder expects a list of Documents as input"):
164
            embedder.run(documents=list_integers_input)
165

166
    def test_run_on_empty_list(self):
167
        embedder = OpenAIDocumentEmbedder(api_key=Secret.from_token("fake-api-key"))
168

169
        empty_list_input = []
170
        result = embedder.run(documents=empty_list_input)
171

172
        assert result["documents"] is not None
173
        assert not result["documents"]  # empty list
174

175
    @pytest.mark.skipif(os.environ.get("OPENAI_API_KEY", "") == "", reason="OPENAI_API_KEY is not set")
176
    @pytest.mark.integration
177
    def test_run(self):
178
        docs = [
179
            Document(content="I love cheese", meta={"topic": "Cuisine"}),
180
            Document(content="A transformer is a deep learning architecture", meta={"topic": "ML"}),
181
        ]
182

183
        model = "text-embedding-ada-002"
184

185
        embedder = OpenAIDocumentEmbedder(model=model, meta_fields_to_embed=["topic"], embedding_separator=" | ")
186

187
        result = embedder.run(documents=docs)
188
        documents_with_embeddings = result["documents"]
189

190
        assert isinstance(documents_with_embeddings, list)
191
        assert len(documents_with_embeddings) == len(docs)
192
        for doc in documents_with_embeddings:
193
            assert isinstance(doc, Document)
194
            assert isinstance(doc.embedding, list)
195
            assert len(doc.embedding) == 1536
196
            assert all(isinstance(x, float) for x in doc.embedding)
197

198
        assert (
199
            "text" in result["meta"]["model"] and "ada" in result["meta"]["model"]
200
        ), "The model name does not contain 'text' and 'ada'"
201

202
        assert result["meta"]["usage"] == {"prompt_tokens": 15, "total_tokens": 15}, "Usage information does not match"
203