Implementation:Vibrantlabsai Ragas LiteLLMEmbeddings

Knowledge Sources	Vibrantlabsai_Ragas
Domains	Embeddings, LiteLLM, LLM Evaluation, Multi-Provider
Last Updated	2026-02-12 00:00 GMT

Overview

LiteLLMEmbeddings provides a universal embedding interface through the LiteLLM library, supporting over 100 embedding models across OpenAI, Azure, Google, Cohere, Anthropic, and other providers.

Description

The LiteLLMEmbeddings class extends BaseRagasEmbedding and uses the LiteLLM library as a universal proxy for embedding API calls. This allows users to switch between embedding providers (OpenAI, Azure, Google, Cohere, and more) without changing their Ragas evaluation code.

Key features include:

• Universal provider support - Access any LiteLLM-supported embedding model through a unified interface using provider-prefixed model names (e.g., "openai/text-embedding-3-small", "cohere/embed-english-v3.0")
• Intelligent batching - Automatically determines optimal batch sizes via the get_optimal_batch_size utility, with manual override capability
• Configurable retry logic - Built-in retry support with configurable max_retries (default 3) and timeout (default 600 seconds)
• Provider-specific parameters - Supports api_key, api_base, and api_version for direct provider configuration, plus arbitrary litellm_params for advanced use cases
• Native async support - Both sync (litellm.embedding) and async (litellm.aembedding) methods are used directly

The _prepare_kwargs helper method centralizes parameter assembly, merging provider configuration, timeout settings, retry logic, and user-supplied keyword arguments into a single call dictionary.

Usage

Use this class when you need a single embedding interface that can work across multiple providers, or when you want to quickly swap between different embedding providers during evaluation experiments. It is ideal for teams that use multiple LLM providers and want consistent Ragas evaluation regardless of the underlying embedding service.

Code Reference

Source Location

• Repository: Vibrantlabsai_Ragas
• File: src/ragas/embeddings/litellm_provider.py

Signature

class LiteLLMEmbeddings(BaseRagasEmbedding):
    PROVIDER_NAME = "litellm"
    REQUIRES_MODEL = True

    def __init__(
        self,
        model: str,
        api_key: Optional[str] = None,
        api_base: Optional[str] = None,
        api_version: Optional[str] = None,
        timeout: int = 600,
        max_retries: int = 3,
        batch_size: Optional[int] = None,
        cache: Optional[CacheInterface] = None,
        **litellm_params: Any,
    ): ...

Import

from ragas.embeddings.litellm_provider import LiteLLMEmbeddings

I/O Contract

Inputs

Name	Type	Required	Description
model	str	Yes	The LiteLLM model identifier, typically provider-prefixed (e.g., "openai/text-embedding-3-small")
api_key	Optional[str]	No	API key for the embedding provider
api_base	Optional[str]	No	Custom API base URL for self-hosted or proxy endpoints
api_version	Optional[str]	No	API version string, used by providers like Azure
timeout	int	No	Timeout in seconds for API calls; defaults to 600
max_retries	int	No	Maximum number of retry attempts for failed API calls; defaults to 3
batch_size	Optional[int]	No	Number of texts to process per batch; auto-determined if not specified
cache	Optional[CacheInterface]	No	Cache backend for storing and retrieving embedding results
**litellm_params	Any	No	Additional keyword arguments passed directly to LiteLLM embedding calls

Outputs

embed_text / aembed_text

Name	Type	Description
return	List[float]	A list of floats representing the embedding vector for a single text

embed_texts / aembed_texts

Name	Type	Description
return	List[List[float]]	A list of embedding vectors, one per input text

Usage Examples

Basic Usage with OpenAI

from ragas.embeddings.litellm_provider import LiteLLMEmbeddings

# Use OpenAI embeddings through LiteLLM
embeddings = LiteLLMEmbeddings(
    model="openai/text-embedding-3-small",
    api_key="sk-your-key-here",
)

vector = embeddings.embed_text("What is retrieval-augmented generation?")
print(len(vector))  # 1536

Azure OpenAI Usage

from ragas.embeddings.litellm_provider import LiteLLMEmbeddings

embeddings = LiteLLMEmbeddings(
    model="azure/my-embedding-deployment",
    api_key="your-azure-key",
    api_base="https://your-resource.openai.azure.com/",
    api_version="2024-02-01",
)

vectors = embeddings.embed_texts([
    "Document about machine learning.",
    "Document about natural language processing.",
])

Async Batch Embedding

import asyncio
from ragas.embeddings.litellm_provider import LiteLLMEmbeddings

embeddings = LiteLLMEmbeddings(
    model="openai/text-embedding-3-small",
    batch_size=50,
    timeout=300,
    max_retries=5,
)

async def embed_corpus(texts):
    return await embeddings.aembed_texts(texts)

texts = ["Text 1", "Text 2", "Text 3"]
result = asyncio.run(embed_corpus(texts))

Related Pages

• Environment:Vibrantlabsai_Ragas_Python_3_9_Core_Environment