Embeddings

Basic Usage

from SimplerLLM.language.embeddings import EmbeddingsLLM, EmbeddingsProvider

embeddings = EmbeddingsLLM.create(provider=EmbeddingsProvider.OPENAI)

vector = embeddings.generate_embeddings("What is machine learning?")
print(len(vector))  # 1536

Providers

Provider	Enum Value	Default Model	Dimensions
OpenAI	EmbeddingsProvider.OPENAI	text-embedding-3-small	1536
Voyage AI	EmbeddingsProvider.VOYAGE	voyage-3	1024
Cohere	EmbeddingsProvider.COHERE	embed-english-v3.0	1024

Batch Embeddings

Pass a list of strings to embed multiple texts in a single API call:

texts = [
    "What is Python?",
    "How does machine learning work?",
    "Explain neural networks"
]

vectors = embeddings.generate_embeddings(texts)
print(len(vectors))     # 3
print(len(vectors[0]))  # 1536

Full Response with Metadata

Set full_response=True to get timing and model info:

response = embeddings.generate_embeddings(
    "What is deep learning?",
    full_response=True
)

print(len(response.generated_embedding))  # 1536
print(f"Model: {response.model}")
print(f"Time: {response.process_time:.2f}s")

Switching Providers

All providers use the same generate_embeddings() method:

from SimplerLLM.language.embeddings import EmbeddingsLLM, EmbeddingsProvider

# OpenAI
embeddings = EmbeddingsLLM.create(provider=EmbeddingsProvider.OPENAI)

# Voyage AI
embeddings = EmbeddingsLLM.create(provider=EmbeddingsProvider.VOYAGE)

# Cohere
embeddings = EmbeddingsLLM.create(provider=EmbeddingsProvider.COHERE)

To use a specific model:

embeddings = EmbeddingsLLM.create(
    provider=EmbeddingsProvider.OPENAI,
    model_name="text-embedding-3-large"  # 3072 dimensions
)

Voyage AI

Voyage AI supports input type optimization and custom dimensions:

embeddings = EmbeddingsLLM.create(provider=EmbeddingsProvider.VOYAGE)

# Optimize for search queries
query_vector = embeddings.generate_embeddings(
    "How does Python async work?",
    input_type="query"
)

# Optimize for documents being indexed
doc_vector = embeddings.generate_embeddings(
    "Python's asyncio module provides infrastructure for writing...",
    input_type="document"
)

Parameter	Type	Description
input_type	str	"query" or "document" — optimizes for retrieval
output_dimension	int	256, 512, 1024, or 2048
output_dtype	str	"float", "int8", "uint8", "binary", "ubinary"

Available models: voyage-3, voyage-3-lite, voyage-code-3, voyage-finance-2.

Cohere

Cohere supports multiple input types for different use cases:

embeddings = EmbeddingsLLM.create(provider=EmbeddingsProvider.COHERE)

# For documents being indexed
doc_vector = embeddings.generate_embeddings(
    "Python is a programming language...",
    input_type="search_document"
)

# For search queries
query_vector = embeddings.generate_embeddings(
    "What is Python?",
    input_type="search_query"
)

Parameter	Type	Default	Description
input_type	str	"search_document"	"search_document", "search_query", "classification", "clustering"
truncate	str	"END"	"START", "END", or "NONE"

Available models: embed-english-v3.0, embed-multilingual-v3.0, embed-v4.0.

Async Usage

import asyncio
from SimplerLLM.language.embeddings import EmbeddingsLLM, EmbeddingsProvider

embeddings = EmbeddingsLLM.create(provider=EmbeddingsProvider.OPENAI)

async def main():
    vector = await embeddings.generate_embeddings_async("What is AI?")
    print(len(vector))

asyncio.run(main())