Ollama was successfully tested on a local computer for loading an embedding model and generate vector embeddings from chunked data. It could handle a load of data in the order of 100k records without breaking and in ~3 hours (Mac Studio with M1 chip). The following actions are needed for integrating it into the pipeline:
[ ] Implement ollama embedder UDF
[ ] Add a CLI option to choose between ollama and gaia node for generating embeddings
Ollama was successfully tested on a local computer for loading an embedding model and generate vector embeddings from chunked data. It could handle a load of data in the order of 100k records without breaking and in ~3 hours (Mac Studio with M1 chip). The following actions are needed for integrating it into the pipeline: