llm-jina project is working well, but responses are raw, and not currently logged.
Before I go on with this plugin I am thinking of making changes to the main llm project.
I want to support other models like, rerankers and classifiers.
I feel a database per model type would grow out of hand quickly. I strongly prefer to treat any request to an ml model as a prompt/response in the responses table.
The main change would be to add a model_type column to the responses table. This could contain: llm, classifier, reranker, etc.
The first issue I guess is that conversation_id doesn't really make sense.
I guess another option would be to have a single additional db for none LLM models. I should look at the embedding db and see if I can reuse that better that the logs db.
llm-jina project is working well, but responses are raw, and not currently logged. Before I go on with this plugin I am thinking of making changes to the main llm project. I want to support other models like, rerankers and classifiers.
I feel a database per model type would grow out of hand quickly. I strongly prefer to treat any request to an ml model as a prompt/response in the responses table.
The main change would be to add a model_type column to the responses table. This could contain: llm, classifier, reranker, etc. The first issue I guess is that conversation_id doesn't really make sense. I guess another option would be to have a single additional db for none LLM models. I should look at the embedding db and see if I can reuse that better that the logs db.