RAG MVP

[chidochipotle]

Adding a minimum minimum viable RAG feature

[chidochipotle]

Hi Simon, added the commits address your previous comments.

Note the last one is unfinished, hope we can finish it in our meeting.
Unfinished tasks:

• available_variables should be added as "source_material" using form_data , but I could not figure out how to do it with form_data.
• A message hint should be added to the UI when files are attach to explain how to add the "context" and "input" variables. (Guess you will do this on your end)
• Testing was minimal see screenshot below:
  

Totally independently:
I don't remember if I reported the bug where there are issues if you attach a file before creating the experiment.
It may not exist anymore.
I had to create an experiment every time I tested, sometimes if I reused old experiments after changing the code there was issues such as getting no answer with the system got in a bad state and started to throw lots of logs.

[snopoke]

Looking good. A few small comments but nothing major.

[snopoke]

Use embeddings from LLM service

Suggested change

	embeddings_model = OpenAIEmbeddings()
	embeddings_model = experiment.get_llm_service().get_openai_embeddings()

[snopoke]

This doesn't need to be a task since it's just called as a normal function

Suggested change

@shared_task(bind=True, base=TaskbadgerTask)

[snopoke]

To execute this as a task you should call .delay:

Suggested change

	store_rag_embedding(experiment)
	store_rag_embedding.delay(experiment.id)

Also we just pass the experiment ID to avoid serialising the whole experiment object

[snopoke]

Rather than taking the experiment object this should take just the ID and fetch the experiment from the DB:

Suggested change

	def store_rag_embedding(self, experiment) -> None:
	def store_rag_embedding(self, experiment_id: int) -> None:
	experiment = Experiment.objects.get(id=experiment_id)

[chidochipotle]

@snopoke, just pushed your fixes.

Tested with a simple pdf file:
RAG_test_file.pdf

The following prompt:

You are an assistant for question-answering tasks. Use the following pieces of retrieved context to answer the question. If you don't know the answer, just say that you don't know. Use three sentences maximum and keep the answer concise.

Question: {input} 

Context: {context} 

Answer:

The following questions:

Hello, you can ask me anything you want about last_test.

What is "Policy objective 1" ?

Policy objective 1 is to reduce newborn, child, and adolescent morbidity and mortality due to preventable communicable diseases.

What is the capital of France ?

I don't know.