Meaning of embedding_mode option

[mikozh]

Method to build a query engine from graph index has parameter named embedding_mode, however I can't find any usage of this parameter. What is the purpose of this parameter? What values does it take and what mechanics are hidden behind it?

Usage example:

llama_index/docs/examples/index_structs/knowledge_graph/KuzuGraphDemo.ipynb

Line 407 in 5a4d246

" embedding_mode=\"hybrid\",\n",

[dosubot[bot]]

🤖

Hey @mikozh, nice to see you diving into the depths of LlamaIndex again! How's the coding journey treating you?

The embedding_mode parameter in the method to build a query engine from the graph index in the LlamaIndex codebase is used to determine the engine for embeddings. This parameter, along with the model parameter, is used in the get_engine function, which utilizes dictionaries like _QUERY_MODE_MODEL_DICT and _TEXT_MODE_MODEL_DICT to map the mode and model to a specific OpenAI embedding model. The mode parameter can take values such as OpenAIEmbeddingMode.SIMILARITY_MODE or OpenAIEmbeddingMode.TEXT_SEARCH_MODE, which are then used along with the model parameter to select the appropriate embedding model, such as "text-similarity-davinci-001" or "text-search-davinci-query-001". You can find more details in the source code.

The possible values that the embedding_mode parameter can take are "similarity" and "text_search". This is mentioned in the legacy code.

The embedding_mode parameter, referred to as retriever_mode in the provided context, interacts with other parameters in the method to build a query engine from the graph index by determining the mode of operation for the retrieval process. This parameter can take one of three values: "keyword", "embedding", or "hybrid", each specifying a different method for finding relevant triplets in the knowledge graph. The interaction of the retriever_mode with other parameters is evident in the _retrieve method, where the logic branches based on the mode to either perform keyword-based retrieval, embedding-based retrieval, or a combination of both. The choice of mode affects how the query engine processes queries, the type of data it retrieves from the graph, and how it constructs the response. You can find more details in the source code.

The embedding_mode parameter impacts the performance of the query engine in the LlamaIndex codebase by determining which OpenAI model is used for generating embeddings. The choice of model affects both the quality and the computational cost of the embeddings. The embedding_mode parameter allows for flexibility in balancing the trade-offs between the quality of embeddings, computational cost, and response time. Choosing the right model based on the specific requirements of the application can significantly impact the performance of the query engine in the LlamaIndex codebase. You can find more details in the source code.

Sources

• llama-index-integrations/embeddings/llama-index-embeddings-openai/llama_index/embeddings/openai/base.py
• llama-index-legacy/llama_index/legacy/embeddings/openai.py
• llama-index-core/llama_index/core/indices/knowledge_graph/retrievers.py

About Dosu

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.