Skip to content

Open Source alternative to Algolia + Pinecone and an Easier-to-Use alternative to ElasticSearch ⚡ 🔍 ✨ Fast, typo tolerant, in-memory fuzzy Search Engine for building delightful search experiences

Use embedding in auto embedding field if already present.

Kishore Nallan3ba9b48eff29
diff --git a/include/validator.h b/include/validator.h
index b4952cf6..d3898dba 100644
--- a/include/validator.h
+++ b/include/validator.h
@@ -71,6 +71,6 @@ public:
static Option<bool> validate_embed_fields(const nlohmann::json& document,
const tsl::htrie_map<char, field>& embedding_fields,
const tsl::htrie_map<char, field> & search_schema,
- const bool& error_if_field_not_found);
+ const bool& is_update);
};
\ No newline at end of file
diff --git a/src/index.cpp b/src/index.cpp
index 69df9301..a87f245c 100644
--- a/src/index.cpp
+++ b/src/index.cpp
@@ -504,6 +504,7 @@ void Index::validate_and_preprocess(Index *index, std::vector<index_record>& ite
index_rec.index_failure(400, e.what());
}
}
+
if(generate_embeddings) {
batch_embed_fields(records_to_embed, embedding_fields, search_schema, remote_embedding_batch_size);
}
@@ -6499,6 +6500,12 @@ void Index::batch_embed_fields(std::vector<index_record*>& records,
if(document == nullptr) {
continue;
}
+
+ if(document->contains(field.name) && !record->is_update) {
+ // embedding already exists (could be a restore from export)
+ continue;
+ }
+
std::string text = indexing_prefix;
const auto& embed_from = field.embed[fields::from].get<std::vector<std::string>>();
for(const auto& field_name : embed_from) {
diff --git a/src/validator.cpp b/src/validator.cpp
index 5fe9bab3..51a7d19c 100644
--- a/src/validator.cpp
+++ b/src/validator.cpp
@@ -654,7 +654,7 @@ Option<uint32_t> validator_t::validate_index_in_memory(nlohmann::json& document,
if(validate_embedding_fields) {
// validate embedding fields
- auto validate_embed_op = validate_embed_fields(document, embedding_fields, search_schema, !is_update);
+ auto validate_embed_op = validate_embed_fields(document, embedding_fields, search_schema, is_update);
if(!validate_embed_op.ok()) {
return Option<>(validate_embed_op.code(), validate_embed_op.error());
}
@@ -667,8 +667,26 @@ Option<uint32_t> validator_t::validate_index_in_memory(nlohmann::json& document,
Option<bool> validator_t::validate_embed_fields(const nlohmann::json& document,
const tsl::htrie_map<char, field>& embedding_fields,
const tsl::htrie_map<char, field> & search_schema,
- const bool& error_if_field_not_found) {
+ const bool& is_update) {
for(const auto& field : embedding_fields) {
+ if(document.contains(field.name) && !is_update) {
+ const auto& field_vec = document[field.name];
+ if(!field_vec.is_array() || field_vec.empty() || !field_vec[0].is_number() ||
+ field_vec.size() != field.num_dim) {
+ return Option<bool>(400, "Field `" + field.name + "` contains an invalid embedding.");
+ }
+
+ auto it = field_vec.begin();
+ while(it != field_vec.end()) {
+ if(!it.value().is_number()) {
+ return Option<bool>(400, "Field `" + field.name + "` contains invalid float values.");
+ }
+ it++;
+ }
+
+ continue;
+ }
+
const auto& embed_from = field.embed[fields::from].get<std::vector<std::string>>();
// flag to check if all fields to embed from are optional and null
bool all_optional_and_null = true;
@@ -679,7 +697,7 @@ Option<bool> validator_t::validate_embed_fields(const nlohmann::json& document,
return Option<bool>(400, "Field `" + field.name + "` has invalid fields to create embeddings from.");
}
if(doc_field_it == document.end()) {
- if(error_if_field_not_found && !schema_field_it->optional) {
+ if(!is_update && !schema_field_it->optional) {
return Option<bool>(400, "Field `" + field_name + "` is needed to create embedding.");
} else {
continue;
diff --git a/test/collection_vector_search_test.cpp b/test/collection_vector_search_test.cpp
index d28d9c03..3ffe288e 100644
--- a/test/collection_vector_search_test.cpp
+++ b/test/collection_vector_search_test.cpp
@@ -1031,4 +1031,74 @@ TEST_F(CollectionVectorTest, EmbedFromOptionalNullField) {
add_op = coll->add(doc.dump());
ASSERT_TRUE(add_op.ok());
-}
\ No newline at end of file
+}
+
+TEST_F(CollectionVectorTest, SkipEmbeddingOpWhenValueExists) {
+ nlohmann::json schema = R"({
+ "name": "objects",
+ "fields": [
+ {"name": "name", "type": "string"},
+ {"name": "embedding", "type":"float[]", "embed":{"from": ["name"], "model_config": {"model_name": "ts/e5-small"}}}
+ ]
+ })"_json;
+
+ TextEmbedderManager::set_model_dir("/tmp/typesense_test/models");
+
+ nlohmann::json model_config = R"({
+ "model_name": "ts/e5-small"
+ })"_json;
+
+ // will be roughly 0.1110895648598671,-0.11710234731435776,-0.5319093465805054, ...
+
+ auto op = collectionManager.create_collection(schema);
+ ASSERT_TRUE(op.ok());
+ Collection* coll = op.get();
+
+ // document with explicit embedding vector
+ nlohmann::json doc;
+ doc["name"] = "FOO";
+
+ std::vector<float> vec;
+ for(size_t i = 0; i < 384; i++) {
+ vec.push_back(0.345);
+ }
+
+ doc["embedding"] = vec;
+
+ auto add_op = coll->add(doc.dump());
+ ASSERT_TRUE(add_op.ok());
+
+ // get the vector back
+ auto res = coll->search("*", {}, "", {}, {}, {0}, 10, 1, FREQUENCY, {true},
+ Index::DROP_TOKENS_THRESHOLD).get();
+
+ // let's check the first few vectors
+ auto stored_vec = res["hits"][0]["document"]["embedding"];
+ ASSERT_NEAR(0.345, stored_vec[0], 0.01);
+ ASSERT_NEAR(0.345, stored_vec[1], 0.01);
+ ASSERT_NEAR(0.345, stored_vec[2], 0.01);
+ ASSERT_NEAR(0.345, stored_vec[3], 0.01);
+ ASSERT_NEAR(0.345, stored_vec[4], 0.01);
+
+ // what happens when vector contains invalid value, like string
+ doc["embedding"] = "foo"; //{0.11, 0.11};
+ add_op = coll->add(doc.dump());
+ ASSERT_FALSE(add_op.ok());
+ ASSERT_EQ("Field `embedding` contains an invalid embedding.", add_op.error());
+
+ // when dims don't match
+ doc["embedding"] = {0.11, 0.11};
+ add_op = coll->add(doc.dump());
+ ASSERT_FALSE(add_op.ok());
+ ASSERT_EQ("Field `embedding` contains an invalid embedding.", add_op.error());
+
+ // invalid array value
+ doc["embedding"].clear();
+ for(size_t i = 0; i < 384; i++) {
+ doc["embedding"].push_back(0.01);
+ }
+ doc["embedding"][5] = "foo";
+ add_op = coll->add(doc.dump());
+ ASSERT_FALSE(add_op.ok());
+ ASSERT_EQ("Field `embedding` contains invalid float values.", add_op.error());
+}