---
language:
- en
- ar
- zh
- fr
- de
- ja
- ko
- es
tags:
- liquid
- lfm2
- edge
- ColBERT
- multi-vector
- PyLate
- sentence-transformers
- sentence-similarity
- feature-extraction
pipeline_tag: sentence-similarity
library_name: PyLate
license: other
license_name: lfm1.0
license_link: LICENSE
metrics:
- MaxSim_accuracy@1
- MaxSim_accuracy@3
- MaxSim_accuracy@5
- MaxSim_accuracy@10
- MaxSim_precision@1
- MaxSim_precision@3
- MaxSim_precision@5
- MaxSim_precision@10
- MaxSim_recall@1
- MaxSim_recall@3
- MaxSim_recall@5
- MaxSim_recall@10
- MaxSim_ndcg@10
- MaxSim_mrr@10
- MaxSim_map@100
new_version: LiquidAI/LFM2.5-ColBERT-350M
model-index:
- name: PyLate
results:
- task:
type: py-late-information-retrieval
name: Py Late Information Retrieval
dataset:
name: NanoClimateFEVER
type: NanoClimateFEVER
metrics:
- type: MaxSim_accuracy@1
value: 0.4
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.52
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.64
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.8
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.4
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.20666666666666667
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.15600000000000003
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.11799999999999997
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.195
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.2733333333333334
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.31566666666666665
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.45899999999999996
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.38664978031372876
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.506095238095238
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.31298040075928324
name: Maxsim Map@100
- task:
type: py-late-information-retrieval
name: Py Late Information Retrieval
dataset:
name: NanoDBPedia
type: NanoDBPedia
metrics:
- type: MaxSim_accuracy@1
value: 0.86
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.92
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.94
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.98
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.86
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.7
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.64
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.5660000000000001
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.12059669442306678
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.2074731836307263
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.28467782707772427
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.4182321427819297
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.7139105590461106
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.8975
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.5749525139913445
name: Maxsim Map@100
- task:
type: py-late-information-retrieval
name: Py Late Information Retrieval
dataset:
name: NanoFEVER
type: NanoFEVER
metrics:
- type: MaxSim_accuracy@1
value: 0.96
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.98
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.98
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.98
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.96
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.3533333333333333
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.21599999999999994
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.10799999999999997
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.8966666666666667
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.9533333333333333
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.96
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.96
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.9494352279872394
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.9666666666666667
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.9396715796806541
name: Maxsim Map@100
- task:
type: py-late-information-retrieval
name: Py Late Information Retrieval
dataset:
name: NanoFiQA2018
type: NanoFiQA2018
metrics:
- type: MaxSim_accuracy@1
value: 0.56
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.78
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.78
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.82
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.56
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.36
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.26
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.14799999999999996
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.34874603174603175
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.5375714285714286
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.584452380952381
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.6361984126984127
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.5909802936451645
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.6625000000000001
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.5325997234510657
name: Maxsim Map@100
- task:
type: py-late-information-retrieval
name: Py Late Information Retrieval
dataset:
name: NanoHotpotQA
type: NanoHotpotQA
metrics:
- type: MaxSim_accuracy@1
value: 0.92
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.98
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 1.0
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 1.0
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.92
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.5599999999999999
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.35999999999999993
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.18799999999999997
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.46
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.84
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.9
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.94
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.8954853297530804
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.9540000000000001
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.8452079490557751
name: Maxsim Map@100
- task:
type: py-late-information-retrieval
name: Py Late Information Retrieval
dataset:
name: NanoMSMARCO
type: NanoMSMARCO
metrics:
- type: MaxSim_accuracy@1
value: 0.58
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.7
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.76
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.82
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.58
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.23333333333333336
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.15200000000000002
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.08199999999999999
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.58
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.7
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.76
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.82
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.6860512766453598
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.6444126984126984
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.6563222143353721
name: Maxsim Map@100
- task:
type: py-late-information-retrieval
name: Py Late Information Retrieval
dataset:
name: NanoNFCorpus
type: NanoNFCorpus
metrics:
- type: MaxSim_accuracy@1
value: 0.5
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.6
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.66
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.7
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.5
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.3933333333333333
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.36400000000000005
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.29
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.06441975062397678
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.10314642255588413
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.1276571146817061
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.15217406670771688
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.37688958487118834
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.5655
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.18401289774934215
name: Maxsim Map@100
- task:
type: py-late-information-retrieval
name: Py Late Information Retrieval
dataset:
name: NanoNQ
type: NanoNQ
metrics:
- type: MaxSim_accuracy@1
value: 0.66
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.78
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.86
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.88
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.66
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.26666666666666666
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.184
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.09599999999999997
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.62
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.72
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.82
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.85
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.7462482063760048
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.7323333333333334
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.7080310107127462
name: Maxsim Map@100
- task:
type: py-late-information-retrieval
name: Py Late Information Retrieval
dataset:
name: NanoQuoraRetrieval
type: NanoQuoraRetrieval
metrics:
- type: MaxSim_accuracy@1
value: 0.8
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.92
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.98
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 1.0
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.8
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.3666666666666666
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.23599999999999993
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.132
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.7106666666666667
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.8813333333333333
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.9346666666666666
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.9793333333333334
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.882106394646597
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.8631666666666666
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.8429613442113442
name: Maxsim Map@100
- task:
type: py-late-information-retrieval
name: Py Late Information Retrieval
dataset:
name: NanoSCIDOCS
type: NanoSCIDOCS
metrics:
- type: MaxSim_accuracy@1
value: 0.5
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.68
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.76
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.86
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.5
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.3466666666666666
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.27599999999999997
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.18599999999999997
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.10566666666666666
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.21366666666666664
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.2826666666666667
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.38066666666666665
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.3835676640413774
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.6130555555555556
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.29746953473534715
name: Maxsim Map@100
- task:
type: py-late-information-retrieval
name: Py Late Information Retrieval
dataset:
name: NanoArguAna
type: NanoArguAna
metrics:
- type: MaxSim_accuracy@1
value: 0.28
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.5
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.7
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.88
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.28
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.16666666666666663
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.14
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.088
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.28
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.5
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.7
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.88
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.550733304467759
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.44868253968253957
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.4511852654234456
name: Maxsim Map@100
- task:
type: py-late-information-retrieval
name: Py Late Information Retrieval
dataset:
name: NanoSciFact
type: NanoSciFact
metrics:
- type: MaxSim_accuracy@1
value: 0.7
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.82
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.88
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.92
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.7
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.29333333333333333
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.19599999999999998
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.10199999999999998
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.675
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.805
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.88
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.91
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.8042579434791977
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.7707142857142858
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.7705943722943722
name: Maxsim Map@100
- task:
type: py-late-information-retrieval
name: Py Late Information Retrieval
dataset:
name: NanoTouche2020
type: NanoTouche2020
metrics:
- type: MaxSim_accuracy@1
value: 0.7959183673469388
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.9795918367346939
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 1.0
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 1.0
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.7959183673469388
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.7619047619047619
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.6897959183673469
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.5489795918367346
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.054536287361574225
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.15427299614340523
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.2308549805548407
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.3471515583210746
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.629619385239878
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.8894557823129251
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.4616787350016793
name: Maxsim Map@100
- task:
type: nano-beir
name: Nano BEIR
dataset:
name: NanoBEIR mean
type: NanoBEIR_mean
metrics:
- type: MaxSim_accuracy@1
value: 0.6550706436420722
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.7815070643642071
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.8415384615384615
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.8953846153846154
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.6550706436420722
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.3852747252747252
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.2976766091051805
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.20407535321821035
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.3931768280118962
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.5299331305821623
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.5985109464051271
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.6717504754237795
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.6612257654240528
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.7318525204953776
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.5828975031847518
name: Maxsim Map@100
---
# LFM2-ColBERT-350M
LFM2-ColBERT-350M is a late interaction retriever with excellent multilingual performance. It allows you to store documents in one language (for example, a product description in English) and retrieve them in many languages with high accuracy.
- LFM2-ColBERT-350M offers **best-in-class accuracy** across different languages.
- Inference speed is **on par with models 2.3 times smaller**, thanks to the efficient LFM2 backbone.
- You can use it as a **drop-in replacement** in your current RAG pipelines to improve performance.
Find more information about LFM2-ColBERT-350M in our [blog post](http://www.liquid.ai/blog/lfm2-colbert-350m-one-model-to-embed-them-all).
> [!NOTE]
> 🚀 Try our demo: https://huggingface.co/spaces/LiquidAI/LFM2-ColBERT
## 📄 Model details
Late interaction retrievers like LFM2-ColBERT-350M are particularly interesting because they preserve much of the **expressivity** of re-rankers while retaining the **efficiency** of bi-encoders.
In practice, they're used to both retrieve documents at scale (like bi-encoders) and rank them at the same time (like rerankers).
We recommend using this model for various RAG use cases, such as:
- **E-commerce**: Find products across many languages with semantic search at scale.
- **On-device semantic search**: Ask questions to your phone in natural language to retrieve files, emails, and notes.
- **Enterprise knowledge assistants**: Retrieve internal legal, financial, and technical documents in different languages.
| Property | [**LFM2-ColBERT-350M**](https://huggingface.co/LiquidAI/LFM2-ColBERT-350M/) |
| --------------------- | ------------------------------ |
| **Total parameters** | 353,322,752 |
| **Layers** | 17 (10 conv + 6 attn + 1 dense)|
| **Vocabulary size** | 64,402 |
| **Training precision**| BF16 |
| **License** | LFM Open License v1.0 |
**Document length:** 512 tokens
**Query length:** 32 tokens
**Output dimensionality:** 128 tokens
**Similarity function:** MaxSim
**Supported languages**: English, Arabic, Chinese, French, German, Japanese, Korean, and Spanish.
```
ColBERT(
(0): Transformer({'max_seq_length': 511, 'do_lower_case': False}) with Transformer model: Lfm2Model
(1): Dense({'in_features': 1024, 'out_features': 128, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity'})
)
```
## 🏃 How to run
### Sentence Transformers
This model can be used with [Sentence Transformers](https://www.sbert.net/) as a multi-vector (ColBERT-style late interaction) retriever via the `MultiVectorEncoder`:
```bash
pip install "sentence-transformers>=6.0.0"
```
```python
from sentence_transformers import MultiVectorEncoder
model = MultiVectorEncoder("LiquidAI/LFM2-ColBERT-350M")
query = "Which planet is known as the Red Planet?"
documents = [
"Venus wird oft als Zwilling der Erde bezeichnet, wegen ihrer ähnlichen Größe.",
"Mars, connue pour son apparence rougeâtre, est souvent appelée la planète rouge.",
"Júpiter es el planeta más grande del sistema solar.",
"Saturno è famoso per i suoi bellissimi anelli.",
]
query_embeddings = model.encode_query(query)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings[0].shape)
# (32, 128) (17, 128)
# MaxSim late-interaction scoring (higher is more relevant)
scores = model.similarity(query_embeddings, document_embeddings)
print(scores)
# tensor([[30.3632, 30.4892, 30.4024, 30.3004]])
```
### PyLate
First, install the PyLate and transformers library:
```bash
pip install -U pylate
```
### Retrieval
Use this model with PyLate to index and retrieve documents. The index uses [FastPLAID](https://github.com/lightonai/fast-plaid) for efficient similarity search.
#### Indexing documents
Load LFM2-ColBERT-350M and initialize the PLAID index, then encode and index your documents:
```python
from pylate import indexes, models, retrieve
# Step 1: Load the ColBERT model
model = models.ColBERT(
model_name_or_path="LiquidAI/LFM2-ColBERT-350M",
)
model.tokenizer.pad_token = model.tokenizer.eos_token
# Step 2: Initialize the PLAID index
index = indexes.PLAID(
index_folder="pylate-index",
index_name="index",
override=True, # This overwrites the existing index if any
)
# Step 3: Encode the documents
documents_ids = ["1", "2", "3"]
documents = ["document 1 text", "document 2 text", "document 3 text"]
documents_embeddings = model.encode(
documents,
batch_size=32,
is_query=False, # Ensure that it is set to False to indicate that these are documents, not queries
show_progress_bar=True,
)
# Step 4: Add document embeddings to the index by providing embeddings and corresponding ids
index.add_documents(
documents_ids=documents_ids,
documents_embeddings=documents_embeddings,
)
```
Note that you do not have to recreate the index and encode the documents every time. Once you have created an index and added the documents, you can re-use the index later by loading it:
```python
# To load an index, simply instantiate it with the correct folder/name and without overriding it
index = indexes.PLAID(
index_folder="pylate-index",
index_name="index",
)
```
#### Retrieving top-k documents for queries
Once the documents are indexed, you can retrieve the top-k most relevant documents for a given set of queries.
To do so, initialize the ColBERT retriever with the index you want to search in, encode the queries and then retrieve the top-k documents to get the top matches ids and relevance scores:
```python
# Step 1: Initialize the ColBERT retriever
retriever = retrieve.ColBERT(index=index)
# Step 2: Encode the queries
queries_embeddings = model.encode(
["query for document 3", "query for document 1"],
batch_size=32,
is_query=True, # # Ensure that it is set to True to indicate that these are queries
show_progress_bar=True,
)
# Step 3: Retrieve top-k documents
scores = retriever.retrieve(
queries_embeddings=queries_embeddings,
k=10, # Retrieve the top 10 matches for each query
)
```
### Reranking
If you only want to use LFM2-ColBERT-350M to perform reranking on top of your first-stage retrieval pipeline without building an index, you can simply use rank function and pass the queries and documents to rerank:
```python
from pylate import rank, models
queries = [
"query A",
"query B",
]
documents = [
["document A", "document B"],
["document 1", "document C", "document B"],
]
documents_ids = [
[1, 2],
[1, 3, 2],
]
model = models.ColBERT(
model_name_or_path="LiquidAI/LFM2-ColBERT-350M",
)
queries_embeddings = model.encode(
queries,
is_query=True,
)
documents_embeddings = model.encode(
documents,
is_query=False,
)
reranked_documents = rank.rerank(
documents_ids=documents_ids,
queries_embeddings=queries_embeddings,
documents_embeddings=documents_embeddings,
)
```
## 📈 Performance
### Accuracy
We extended the NanoBEIR benchmark to include Japanese and Korean languages. We open-sourced this dataset on Hugging Face at [LiquidAI/nanobeir-multilingual-extended](https://huggingface.co/datasets/LiquidAI/nanobeir-multilingual-extended) for reproducibility.
On this NanoBEIR benchmark, LFM2-ColBERT-350M displays significantly stronger multilingual capabilities (especially in German, Arabic, Korean, and Japanese) while maintaining English performance.
Even more interestingly, LFM2-ColBERT-350M is an excellent cross-lingual retriever. This means that it is capable of retrieving documents based on queries from other languages. This is ideal for client-facing applications, like in e-commerce, where a description might be in English but the query is in another language.
LFM2-ColBERT-350M works especially well for English, French, Spanish, Italian, Portuguese, and German, as shown with these NDCG@10 scores on NanoBEIR:
| Doc / Query |
AR |
DE |
EN |
ES |
FR |
IT |
JA |
KO |
PT |
AVG |
| AR |
0.490 |
0.288 |
0.339 |
0.303 |
0.304 |
0.286 |
0.357 |
0.338 |
0.291 |
33.30% |
| DE |
0.383 |
0.563 |
0.547 |
0.498 |
0.502 |
0.489 |
0.424 |
0.368 |
0.486 |
47.33% |
| EN |
0.416 |
0.554 |
0.661 |
0.553 |
0.551 |
0.522 |
0.477 |
0.395 |
0.535 |
51.82% |
| ES |
0.412 |
0.514 |
0.578 |
0.563 |
0.547 |
0.529 |
0.436 |
0.394 |
0.547 |
50.21% |
| FR |
0.408 |
0.527 |
0.573 |
0.552 |
0.564 |
0.537 |
0.450 |
0.388 |
0.549 |
50.53% |
| IT |
0.395 |
0.512 |
0.554 |
0.535 |
0.535 |
0.543 |
0.439 |
0.386 |
0.529 |
49.20% |
| JA |
0.375 |
0.365 |
0.409 |
0.358 |
0.345 |
0.337 |
0.557 |
0.491 |
0.330 |
39.63% |
| KO |
0.326 |
0.274 |
0.310 |
0.282 |
0.265 |
0.266 |
0.440 |
0.527 |
0.271 |
32.89% |
| PT |
0.402 |
0.499 |
0.558 |
0.545 |
0.528 |
0.529 |
0.436 |
0.382 |
0.547 |
49.17% |
| AVG |
40.07% |
45.51% |
50.32% |
46.54% |
46.00% |
44.86% |
44.62% |
40.78% |
45.38% |
|
In comparison, GTE-ModernColBERT-v1 consistently gets lower scores when documents and queries are not in the same language:
| Doc / Query |
AR |
DE |
EN |
ES |
FR |
IT |
JA |
KO |
PT |
AVG |
| AR |
0.309 |
0.089 |
0.107 |
0.089 |
0.094 |
0.092 |
0.070 |
0.049 |
0.087 |
10.96% |
| DE |
0.039 |
0.499 |
0.454 |
0.362 |
0.393 |
0.367 |
0.133 |
0.061 |
0.361 |
29.65% |
| EN |
0.042 |
0.408 |
0.680 |
0.446 |
0.484 |
0.420 |
0.167 |
0.073 |
0.438 |
35.08% |
| ES |
0.044 |
0.360 |
0.485 |
0.525 |
0.465 |
0.437 |
0.149 |
0.061 |
0.487 |
33.48% |
| FR |
0.044 |
0.381 |
0.505 |
0.455 |
0.546 |
0.428 |
0.136 |
0.057 |
0.467 |
33.35% |
| IT |
0.043 |
0.369 |
0.449 |
0.446 |
0.451 |
0.516 |
0.143 |
0.054 |
0.448 |
32.36% |
| JA |
0.031 |
0.169 |
0.250 |
0.172 |
0.177 |
0.169 |
0.459 |
0.059 |
0.165 |
18.35% |
| KO |
0.030 |
0.134 |
0.169 |
0.127 |
0.133 |
0.125 |
0.090 |
0.368 |
0.124 |
14.45% |
| PT |
0.043 |
0.368 |
0.479 |
0.492 |
0.467 |
0.448 |
0.138 |
0.062 |
0.530 |
33.63% |
| AVG |
6.94% |
30.84% |
39.75% |
34.59% |
35.68% |
33.35% |
16.53% |
9.37% |
34.24% |
|
This makes retrieval a lot more reliable and can replace architectures with multiple models with a single, unified retriever.
### Inference speed
Despite being more than twice as big, LFM2-ColBERT-350M demonstrates throughput performance on par with GTE-ModernColBERT-v1 for query and document encoding across various batch sizes.
Query encoding was evaluated using realistic query patterns from datasets like MS MARCO and Natural Questions.
Document encoding was measured on realistic documents with varying lengths and domains.
## 📬 Contact
- Got questions or want to connect? [Join our Discord community](https://discord.com/invite/liquid-ai)
- If you are interested in custom solutions with edge deployment, please contact [our sales team](https://www.liquid.ai/contact).
## Citation
```
@article{liquidai2025lfm2,
title={LFM2 Technical Report},
author={Liquid AI},
journal={arXiv preprint arXiv:2511.23404},
year={2025}
}
```
```bibtex
@misc{PyLate,
title={PyLate: Flexible Training and Retrieval for Late Interaction Models},
author={Chaffin, Antoine and Sourty, Raphaël},
url={https://github.com/lightonai/pylate},
year={2024}
}
```