Sample YAML file guide¶

We first categorized them into English and Korean.

Korean consists of only modules that support Korean, such as Korean prompts and Korean rerankers for modules that use LLMs.

We have divided the sample YAML file into three main parts.

I’ll explain how it was divided based on the full.yaml in each folder.

GPU + API¶

API: Four Reranker Modules that use APIs except for the Embedding Model and LLM
GPU (recommend): The Reranker module and the LongLLM Lingua module in the Compressor Node, which can take a long time if you don’t use the GPU and only use the CPU.

JINA AI: You can get API Key here

export JINAAI_API_KEY=your_jina_api_key

Cohere: You can get API Key here

export COHERE_API_KEY=your_cohere_api_key

VoyageAI:You can get API Key here

export VOYAGE_API_KEY=your_voyageai_api_key

Mixbread AI: You can get API Key here

export MXBAI_API_KEY=your_mixedbread_api_key

This excludes the four Reranker modules, which require a separate API key to use.

Exclude API and GPU Recommend from all.

Each folder is again organized into Full, Half, Compact, and even Simple for non-GPUs.

Exclude Query Expansion Node from Full

Exclude Passage Compressor Node from Half

This is the YAML file for the Naive RAG that leads to Retrieval → Prompt Maker → Generator.

I only put it in the Non-GPU folder because it is the only Node that runs in a non-GPU environment.