-
Notifications
You must be signed in to change notification settings - Fork 16
/
Lucene.kt
161 lines (140 loc) · 6.13 KB
/
Lucene.kt
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
package com.xebia.functional.xef.store
import ai.xef.openai.OpenAIModel
import arrow.atomic.AtomicInt
import com.xebia.functional.openai.apis.EmbeddingsApi
import com.xebia.functional.openai.models.ChatCompletionRole
import com.xebia.functional.openai.models.CreateEmbeddingRequestModel
import com.xebia.functional.openai.models.Embedding
import com.xebia.functional.xef.llm.embedQuery
import com.xebia.functional.xef.llm.models.modelType
import org.apache.lucene.analysis.standard.StandardAnalyzer
import org.apache.lucene.document.*
import org.apache.lucene.index.DirectoryReader
import org.apache.lucene.index.IndexWriter
import org.apache.lucene.index.IndexWriterConfig
import org.apache.lucene.index.VectorSimilarityFunction
import org.apache.lucene.queries.mlt.MoreLikeThis
import org.apache.lucene.search.*
import org.apache.lucene.store.Directory
import org.apache.lucene.store.MMapDirectory
import java.io.StringReader
import java.nio.file.Path
open class Lucene(
private val writer: IndexWriter,
private val embeddings: EmbeddingsApi?,
private val similarity: VectorSimilarityFunction = VectorSimilarityFunction.EUCLIDEAN,
private val embeddingAIModel: OpenAIModel<CreateEmbeddingRequestModel>
) : VectorStore, AutoCloseable {
override val indexValue: AtomicInt = AtomicInt(0)
override fun updateIndexByConversationId(conversationId: ConversationId) {
getMemoryByConversationId(conversationId).firstOrNull()?.let { indexValue.set(it.index) }
}
override suspend fun addMemories(memories: List<Memory>) {
memories.forEach {
val doc =
Document().apply {
add(TextField("conversationId", it.conversationId.value, Field.Store.YES))
add(TextField("content", it.content.asRequestMessage().contentAsString(), Field.Store.YES))
add(TextField("role", it.content.role.name.lowercase(), Field.Store.YES))
add(IntField("index", it.index, Field.Store.YES))
}
writer.addDocument(doc)
}
writer.commit()
}
override suspend fun <T> memories(
model: OpenAIModel<T>, conversationId: ConversationId, limitTokens: Int
): List<Memory> =
getMemoryByConversationId(conversationId).reduceByLimitToken(model.modelType(), limitTokens).reversed()
override suspend fun addTexts(texts: List<String>) {
texts.forEach {
val embedding = embeddings?.embedQuery(text = it, embeddingRequestModel = embeddingAIModel)
val doc =
Document().apply {
add(TextField("contents", it, Field.Store.YES))
if (embedding != null) add(KnnFloatVectorField("embedding", embedding.toFloatArray(), similarity))
}
writer.addDocument(doc)
}
writer.commit()
}
override suspend fun similaritySearch(query: String, limit: Int): List<String> {
val reader = DirectoryReader.open(writer)
val mlt = MoreLikeThis(reader)
mlt.analyzer = StandardAnalyzer()
mlt.minTermFreq = 1
mlt.minDocFreq = 1
mlt.minWordLen = 3
val luceneQuery = mlt.like("contents", StringReader(query))
val searcher = IndexSearcher(reader)
return IndexSearcher(reader).search(luceneQuery, limit).extract(searcher)
}
override suspend fun similaritySearchByVector(embedding: Embedding, limit: Int): List<String> {
requireNotNull(embeddings) { "no embeddings were computed for this model" }
val luceneQuery = KnnFloatVectorQuery("embedding", embedding.embedding.map { it.toFloat() }.toFloatArray(), limit)
val searcher = IndexSearcher(DirectoryReader.open(writer))
return searcher.search(luceneQuery, limit).extract(searcher)
}
private fun List<ScoreDoc>.extractMemory(searcher: IndexSearcher): List<Memory> =
map {
val doc = searcher.storedFields().document(it.doc)
val role = ChatCompletionRole.valueOf(doc.get("role").lowercase())
val content = doc.get("content")
Memory(
conversationId = ConversationId(doc.get("conversationId")),
content = memorizedMessage(role, content),
index = doc.get("index").toInt()
)
}
private fun TopDocs.extract(searcher: IndexSearcher): List<String> =
scoreDocs.map {
searcher.storedFields().document(it.doc).get("contents")
}
override fun close() {
writer.close()
}
private fun getMemoryByConversationId(conversationId: ConversationId): List<Memory> {
val reader = DirectoryReader.open(writer)
val mlt = MoreLikeThis(reader)
mlt.analyzer = StandardAnalyzer()
mlt.minTermFreq = 1
mlt.minDocFreq = 1
mlt.minWordLen = 3
val sort = Sort(SortField("index", SortField.Type.LONG, true))
val luceneQuery = mlt.like("conversationId", StringReader(conversationId.value))
val searcher = IndexSearcher(reader)
val docs = IndexSearcher(reader).search(luceneQuery, reader.numDocs(), sort)
return docs.scoreDocs.toList().extractMemory(searcher)
}
}
class DirectoryLucene(
private val directory: Directory,
writerConfig: IndexWriterConfig = IndexWriterConfig(),
embeddings: EmbeddingsApi?,
embeddingAIModel: OpenAIModel<CreateEmbeddingRequestModel>,
similarity: VectorSimilarityFunction = VectorSimilarityFunction.EUCLIDEAN
) : Lucene(IndexWriter(directory, writerConfig), embeddings, similarity, embeddingAIModel) {
override fun close() {
super.close()
directory.close()
}
}
@JvmOverloads
fun InMemoryLucene(
path: Path,
writerConfig: IndexWriterConfig = IndexWriterConfig(),
embeddings: EmbeddingsApi?,
embeddingAIModel: OpenAIModel<CreateEmbeddingRequestModel>,
similarity: VectorSimilarityFunction = VectorSimilarityFunction.EUCLIDEAN
): DirectoryLucene = DirectoryLucene(MMapDirectory(path), writerConfig, embeddings, embeddingAIModel, similarity)
@JvmOverloads
fun InMemoryLuceneBuilder(
path: Path,
useAIEmbeddings: Boolean = true,
writerConfig: IndexWriterConfig = IndexWriterConfig(),
embeddingAIModel: OpenAIModel<CreateEmbeddingRequestModel>,
similarity: VectorSimilarityFunction = VectorSimilarityFunction.EUCLIDEAN
): (EmbeddingsApi) -> DirectoryLucene = { embeddings ->
InMemoryLucene(path, writerConfig, embeddings.takeIf { useAIEmbeddings }, embeddingAIModel, similarity)
}
fun List<Embedding>.toFloatArray(): FloatArray = flatMap { it.embedding.map { it.toFloat() } }.toFloatArray()