from langchain_community.embeddings import HuggingFaceEmbeddings, HuggingFaceInferenceAPIEmbeddings
from langchain_community.vectorstores import Qdrant
from langchain_qdrant import FastEmbedSparse, RetrievalMode
# get the device to be used eithe gpu or cpu
device = 'cuda' if cuda.is_available() else 'cpu'


def hybrid_embed_chunks(chunks):
    """
    takes the chunks and does the hybrid embedding for the list of chunks
    """
    embeddings = HuggingFaceEmbeddings(
        model_kwargs = {'device': device},
        encode_kwargs = {'normalize_embeddings': True},
        model_name='BAAI/bge-m3'
    )
    sparse_embeddings = FastEmbedSparse(model_name="Qdrant/bm25")
    # placeholder for collection
    print("starting embedding")
    #qdrant_collections = {}
    Qdrant.from_documents(
                chunks,
                embeddings,
                sparse_embeddings = sparse_embeddings,
                path="/data/local_qdrant",
                collection_name='giz_worldwide',
                retrieval_mode=RetrievalMode.HYBRID,
            )    
            
    print(qdrant_collections)
    print("vector embeddings done")