Heuristic:Vibrantlabsai Ragas Warning Deprecated Legacy LLM Wrappers

Knowledge Sources	Vibrantlabsai_Ragas
Domains	Deprecation, Migration, LLM_Integration
Last Updated	2026-02-12 00:00 GMT

Overview

DEPRECATION WARNING: The legacy framework wrapper classes LangchainLLMWrapper, LlamaIndexLLMWrapper, LangchainEmbeddingsWrapper, and LlamaIndexEmbeddingsWrapper are deprecated. Users should migrate to llm_factory() and modern embedding providers.

Description

Ragas originally used wrapper classes to bridge LangChain and LlamaIndex LLM/embedding interfaces into its own abstraction. These wrappers are now deprecated in favor of:

llm_factory() with direct OpenAI/provider clients for LLMs
Modern embedding providers (OpenAIEmbeddings, GoogleEmbeddings, HuggingFaceEmbeddings, LiteLLMEmbeddings) for embeddings

Additionally, the evaluate() and aevaluate() functions are deprecated in favor of the @experiment decorator pattern.

Usage

Apply this heuristic when encountering any of the deprecated wrapper classes or the legacy evaluate functions. Migrate to the modern API.

The Insight (Rule of Thumb)

Action (LLMs): Replace LangchainLLMWrapper(chat_model) with llm_factory("openai:gpt-4o") or llm_factory(model="model_name", provider="provider")
Action (Embeddings): Replace LangchainEmbeddingsWrapper(embedder) with OpenAIEmbeddings() or other modern providers
Action (Evaluation): Replace evaluate(dataset, metrics) with the @experiment decorator pattern
Trade-off: Modern API requires explicit provider selection rather than wrapping existing framework objects; may require refactoring initialization code

Reasoning

The wrapper approach added unnecessary indirection and made it harder to support structured outputs and advanced LLM features. The llm_factory() approach uses instructor for structured output generation, providing more reliable JSON parsing and type-safe responses. Modern embedding providers implement BaseRagasEmbeddings directly, offering better control over batching, caching, and async behavior.

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment