From afd6a9f82716d3433873682a315d4fc0dae46931 Mon Sep 17 00:00:00 2001 From: Patrick von Platen Date: Fri, 18 Sep 2020 11:41:12 +0200 Subject: [PATCH] Create README.md --- .../facebook/rag-sequence-nq/README.md | 25 +++++++++++++++++++ 1 file changed, 25 insertions(+) create mode 100644 model_cards/facebook/rag-sequence-nq/README.md diff --git a/model_cards/facebook/rag-sequence-nq/README.md b/model_cards/facebook/rag-sequence-nq/README.md new file mode 100644 index 0000000000..13ab4f64c5 --- /dev/null +++ b/model_cards/facebook/rag-sequence-nq/README.md @@ -0,0 +1,25 @@ +## RAG + +This is the RAG-Sequence Model of the the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/pdf/2005.11401.pdf) +by Aleksandra Piktus et al. + +## Usage: + +```python + +from transformers import RagTokenizer, RagRetriever, RagSequenceForGeneration + +tokenizer = RagTokenizer.from_pretrained("facebook/rag-token-nq") +retriever = RagRetriever.from_pretrained("facebook/rag-token-nq", index_name="exact", use_dummy_dataset=True) +model = RagSequenceForGeneration.from_pretrained("facebook/rag-token-nq", retriever=retriever) + +input_dict = tokenizer.prepare_seq2seq_batch("How many people live in Paris?", "In Paris, there are 10 million people.", return_tensors="pt") +outputs = model(input_ids=input_dict["input_ids"], labels=input_dict["labels"]) + +# outputs.loss should give 76.2978 + +generated = model.generate(input_ids=input_dict["input_ids"], num_beams=4) +generated_string = tokenizer.batch_decode(generated, skip_special_tokens=True) + +# generated_string should give 270,000,000 -> not quite correct the answer, but it also only uses a dummy index +```