metadata
language:
- de
tags:
- ColBERT
- multi-vector
- PyLate
- sentence-transformers
- sentence-similarity
pipeline_tag: sentence-similarity
library_name: PyLate
datasets:
- samheym/ger-dpr-collection
base_model:
- deepset/gbert-base
Model Overview
GerColBERT is a ColBERT-based retrieval model trained on German text. It is designed for efficient late interaction-based retrieval while maintaining high-quality ranking performance. Training Configuration
- Base Model: deepset/gbert-base
- Training Dataset: samheym/ger-dpr-collection
- Dataset: 10% of randomly selected triples from the final dataset
- Vector Length: 128
- Maximum Document Length: 256 Tokens
- Batch Size: 50
- Training Steps: 80,000
- Gradient Accumulation: 1 step
- Learning Rate: 5 × 10⁻⁶
- Optimizer: AdamW
- In-Batch Negatives: Included
Usage
Sentence Transformers
This model can be used with Sentence Transformers as a multi-vector (ColBERT-style late interaction) retriever via the MultiVectorEncoder:
pip install "sentence-transformers>=6.0.0"
from sentence_transformers import MultiVectorEncoder
model = MultiVectorEncoder("samheym/GerColBERT")
query = "Welcher Planet ist als der Rote Planet bekannt?"
documents = [
"Venus wird wegen ihrer ähnlichen Größe und Nähe oft als Zwilling der Erde bezeichnet.",
"Mars, bekannt für sein rötliches Aussehen, wird oft als der Rote Planet bezeichnet.",
"Jupiter, der größte Planet des Sonnensystems, hat einen markanten roten Fleck.",
"Saturn, berühmt für seine Ringe, wird manchmal mit dem Roten Planeten verwechselt.",
]
query_embeddings = model.encode_query(query)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings[0].shape)
# (32, 128) (18, 128)
# MaxSim late-interaction scoring (higher is more relevant)
scores = model.similarity(query_embeddings, document_embeddings)
print(scores)
# tensor([[13.8994, 27.9157, 21.6549, 22.7871]])
PyLate
First install the PyLate library:
pip install -U pylate
Retrieval
PyLate provides a streamlined interface to index and retrieve documents using ColBERT models. The index leverages the Voyager HNSW index to efficiently handle document embeddings and enable fast retrieval.
from pylate import indexes, models, retrieve
# Step 1: Load the ColBERT model
model = models.ColBERT(
model_name_or_path=samheym/GerColBERT,
)