DataScience-UIBK

/

Reason-mxbai-colbert-v0.1-32m

like 2

Follow

DataScienceUIBK 9

Hello!

As of Sentence Transformers v6.0.0, this checkpoint loads directly as a multi-vector (ColBERT-style late interaction) retriever through the new MultiVectorEncoder. This PR adds a Sentence Transformers usage section to the model card and the multi-vector tag. The weights and the existing usage are untouched. The example query keeps the instruction plus Query: format from your PyLate example.

from sentence_transformers import MultiVectorEncoder

model = MultiVectorEncoder("DataScience-UIBK/Reason-mxbai-colbert-v0.1-32m")

query = "Given a Psychology post, retrieve relevant passages that help answer the post.\nQuery: why do I procrastinate?"
documents = [
    "Procrastination is often driven by difficulty regulating negative emotions around a task, not laziness, since delaying provides short term relief.",
    "The hippocampus plays a central role in consolidating short term memories into long term storage.",
]

query_embeddings = model.encode_query(query)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings[0].shape)
# torch.Size([28, 128]) torch.Size([31, 128])

# MaxSim late-interaction scoring (higher is more relevant)
scores = model.similarity(query_embeddings, document_embeddings)
print(scores)
# tensor([[25.2429, 24.4635]], device='cuda:0')

Verified against a PyLate reference: the snippet reproduces exactly, and the token embeddings match with per-token cosine similarity above 0.999 and matching MaxSim scores. For reference, loaded through this integration the model scores 0.6486 mean nDCG@10 on NanoBEIR.

• Tom Aarsen