Create Embeddings Generate embedding vectors that represent the semantic meaning of input text.
require "openai" client = OpenAI :: Client . new response = client. embeddings . create ( input: "The quick brown fox jumped over the lazy dog" , model: "text-embedding-3-small" ) # Access the embedding vector embedding = response. data . first . embedding puts "Embedding dimensions: #{ embedding. length } " puts "First 5 values: #{ embedding. take ( 5 ) } "
Embedding Models OpenAI offers several embedding models with different capabilities and sizes.
text-embedding-3-small
text-embedding-3-large
text-embedding-ada-002
# Fast and efficient for most use cases response = client. embeddings . create ( input: "Sample text" , model: "text-embedding-3-small" )
Batch Embeddings Create embeddings for multiple texts in a single API call for better efficiency.
texts = [ "The quick brown fox jumps over the lazy dog" , "Machine learning is a subset of artificial intelligence" , "Ruby is a dynamic, open source programming language" ] response = client. embeddings . create ( input: texts, model: "text-embedding-3-small" ) # Process each embedding response. data . each_with_index do | item , i | puts "Text #{ i + 1 } : #{ texts[i] } " puts "Embedding index: #{ item. index } " puts "Dimensions: #{ item. embedding . length } \n\n " end
Batch processing is more efficient than making individual API calls for each text.
Custom Dimensions Reduce embedding dimensions for text-embedding-3 models to save storage and improve performance.
response = client. embeddings . create ( input: "Sample text for dimension reduction" , model: "text-embedding-3-small" , dimensions: 512 # Default is 1536 ) embedding = response. data . first . embedding puts "Reduced dimensions: #{ embedding. length } " # 512
Dimension reduction is only supported in text-embedding-3-small and text-embedding-3-large models.
Choose between float and base64 encoding formats.
response = client. embeddings . create ( input: "Sample text" , model: "text-embedding-3-small" , encoding_format: "float" ) # Returns array of floating point numbers embedding = response. data . first . embedding puts embedding. class # Array puts embedding. first . class # Float
require "base64" response = client. embeddings . create ( input: "Sample text" , model: "text-embedding-3-small" , encoding_format: "base64" ) # Returns base64-encoded string encoded = response. data . first . embedding # Decode if needed decoded = Base64 . decode64 (encoded)
Similarity Search Use embeddings to find semantically similar texts by comparing vector distances.
require "matrix" # Calculate cosine similarity between two embeddings def cosine_similarity ( a , b ) dot_product = a. zip (b). map { | x , y | x * y }. sum magnitude_a = Math . sqrt (a. map { | x | x ** 2 }. sum ) magnitude_b = Math . sqrt (b. map { | x | x ** 2 }. sum ) dot_product / (magnitude_a * magnitude_b) end # Create embeddings for query and documents query = "What is machine learning?" documents = [ "Machine learning is a type of AI" , "Ruby is a programming language" , "Deep learning uses neural networks" ] all_texts = [query] + documents response = client. embeddings . create ( input: all_texts, model: "text-embedding-3-small" ) query_embedding = response. data [ 0 ]. embedding doc_embeddings = response. data [ 1 ..]. map ( & :embedding ) # Find most similar document similarities = doc_embeddings. map do | doc_emb | cosine_similarity (query_embedding, doc_emb) end most_similar_idx = similarities. each_with_index . max_by { | sim , _ | sim }[ 1 ] puts "Most similar document: #{ documents[most_similar_idx] } " puts "Similarity score: #{ similarities[most_similar_idx] } "
Response Structure Understand the structure of the embedding response.
response = client. embeddings . create ( input: "Sample text" , model: "text-embedding-3-small" ) # Response object structure puts response. data # Array of embedding objects puts response. model # Model used puts response. object # "list" puts response. usage . prompt_tokens # Tokens used puts response. usage . total_tokens # Total tokens # Individual embedding structure embedding_obj = response. data . first puts embedding_obj. embedding # Vector of floats puts embedding_obj. index # Index in array (0-based) puts embedding_obj. object # "embedding"
Token Arrays Pass tokenized input directly as integer arrays.
# Single token array response = client. embeddings . create ( input: [ 123 , 456 , 789 ], # Array of token IDs model: "text-embedding-3-small" ) # Multiple token arrays response = client. embeddings . create ( input: [ [ 123 , 456 ], [ 789 , 012 ] ], model: "text-embedding-3-small" )
Token arrays must not exceed 2048 dimensions for all embedding models, and the total token count across all inputs is subject to API limits.