Why RAG grounds LLMs in real data in Prompt Engineering / GenAI - Test Your Understanding

Practice - 5 Tasks

1fill in blank

easy

Complete the code to import the RAG model class from transformers.

Prompt Engineering / GenAI

from transformers import [1]

Drag options to blanks, or click blank then click option'

ARagTokenizer

BRagSequenceForGeneration

CRagRetriever

DRagModel

Attempts:

3 left

💡 Hint

Common Mistakes

Choosing tokenizer or retriever classes instead of the main model.

Confusing RagSequenceForGeneration with RagModel.

The RagModel class is the core RAG model that combines retrieval and generation.

2fill in blank

medium

Complete the code to initialize the retriever for RAG using a dataset.

Prompt Engineering / GenAI

retriever = RagRetriever.from_pretrained('facebook/rag-token-base', index_name=[1])

Drag options to blanks, or click blank then click option'

A'exact'

B'custom'

C'legacy'

D'default'

Attempts:

3 left

💡 Hint

Common Mistakes

Using 'custom' or 'default' which are not valid index names.

Confusing 'legacy' with the correct retrieval method.

The exact index_name tells the retriever to use exact matching for retrieval.

3fill in blank

hard

Fix the error in the code to generate an answer grounded on retrieved documents.

Prompt Engineering / GenAI

outputs = model.generate(input_ids, context_doc_ids=[1])

Drag options to blanks, or click blank then click option'

Aretrieved_docs

Bretrieved_doc_embeds

Cretrieved_doc_ids

Dretrieved_doc_texts

Attempts:

3 left

💡 Hint

Common Mistakes

Passing raw document texts instead of IDs.

Passing embeddings which are not accepted here.

The context_doc_ids argument expects the IDs of retrieved documents, not the raw documents or embeddings.

4fill in blank

hard

Fill both blanks to create a RAG model with retriever and generate output.

Prompt Engineering / GenAI

model = RagSequenceForGeneration.from_pretrained('facebook/rag-token-base', retriever=[1])
outputs = model.generate([2])

Drag options to blanks, or click blank then click option'

Aretriever

Binput_ids

Cinput_texts

Dtokenizer

Attempts:

3 left

💡 Hint

Common Mistakes

Passing raw texts instead of token IDs to generate.

Not passing the retriever when creating the model.

The model needs the retriever object to fetch documents, and input_ids to generate answers.

5fill in blank

hard

Fill all three blanks to tokenize input, retrieve docs, and generate grounded output.

Prompt Engineering / GenAI

inputs = tokenizer([1], return_tensors='pt')
retrieved_docs = retriever.retrieve([2])
outputs = model.generate(input_ids=inputs['input_ids'], context_doc_ids=[3])

Drag options to blanks, or click blank then click option'

A'What is RAG?'

Binputs['input_ids']

Cretrieved_docs['doc_ids']

Dretrieved_docs['texts']

Attempts:

3 left

💡 Hint

Common Mistakes

Using raw texts instead of token IDs for retrieval or generation.

Mixing up document texts and document IDs.

First, we tokenize the question string. Then we retrieve documents using the token IDs. Finally, we generate output using input IDs and retrieved document IDs.