Need help in loading the dense retriever from local

[rajeevmajumdar1981]

hi
i have followed the tutorial https://haystack.deepset.ai/tutorials/18_gpl for Generative Pseudo Labeling for Domain Adaptation in which at the end the retriever is saved in local disk. So i was trying to load that retreiver back to create a pipeline with retriever and reader nodes to answer questions but while running the pipeline it gave me below error . while training with model also i never used open AI models then why this is asking API key for openAI

Exception: Exception while running node 'Retriever': OpenAI API key is not set. You can set it via the api_key parameter of the EmbeddingRetriever.

The code i used to load and query

FAISS_INDEX_DOCUMENT = Path(os.getcwd()) / 'document_faiss'
stored_retriever = 'adapted_retriever' #'retriever.pt'
pipeline = Pipeline()
reader = FARMReader("deepset/minilm-uncased-squad2")
def initialise():

if os.path.exists(FAISS_INDEX_DOCUMENT):
    document_store = FAISSDocumentStore.load(FAISS_INDEX_DOCUMENT)
    retriever = EmbeddingRetriever(document_store=document_store, 
                                    use_gpu=False, 
                                    model_format="sentence_transformers", 
                                    embedding_model=stored_retriever, 
                                    top_k=3)
else:
    print ("Error")
    sys.exit()

pipeline.add_node(component=retriever, name='Retriever', inputs=['Query'])  
pipeline.add_node(component=reader, name='Reader', inputs=['Retriever']) 

def query():
question = "how to design wings"
res = pipeline.run(query=question)
print_documents(res, max_text_len=100, print_name=True, print_meta=True)

initialise()
query()

[julian-risch]

Hi @rajeevmajumdar1981 this error should only occur if the EmbeddingRetriever that was loaded had the model_format set to openai. In that case, an _OpenAIEmbeddingEncoder is used internally, which requires an OpenAI API key. I will try to reproduce the error with the colab notebook from the GPL tutorial.

[julian-risch]

I was able to reproduce the issue you reported by running

retriever2 = EmbeddingRetriever(
    document_store=document_store,
    embedding_model="adapted_retriever",
    max_seq_len=max_seq_length,
    progress_bar=False,
)
retriever2.retrieve(query="test")

at the end of the GPL Tutorial.

[julian-risch]

but if I set the model_format sentence_transformers, then the retriever is loaded correctly:

retriever2 = EmbeddingRetriever(
    document_store=document_store,
    embedding_model="adapted_retriever",
    model_format="sentence_transformers",
    max_seq_len=max_seq_length,
    progress_bar=False,
)
retriever2.retrieve(query="test")

@rajeevmajumdar1981 Could you please try that again?