---
language:
- en
tags:
- ColBERT
- PyLate
- sentence-transformers
- sentence-similarity
- feature-extraction
- generated_from_trainer
- dataset_size:640000
- loss:Distillation
base_model: NeuML/bert-hash-pico
datasets:
- lightonai/ms-marco-en-bge-gemma
pipeline_tag: sentence-similarity
library_name: PyLate
license: apache-2.0
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
model-index:
- name: ColBERT MUVERA Pico
results:
- task:
type: py-late-information-retrieval
name: Py Late Information Retrieval
dataset:
name: NanoClimateFEVER
type: NanoClimateFEVER
metrics:
- type: MaxSim_accuracy@1
value: 0.22
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.32
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.4
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.54
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.22
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.11333333333333333
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.09200000000000001
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.062
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.125
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.18
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.22
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.2723333333333333
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.22523375350232466
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.2918015873015873
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.17799142530014567
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: 0.68
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.82
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.86
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.9
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.68
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.5733333333333333
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.49999999999999994
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.47
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.054334623940057496
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.13885343313592602
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.19014904910202987
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.32073944983710506
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.5579484125370805
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.7584126984126983
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.42613646438564345
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: 0.68
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.78
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: 0.68
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.27333333333333326
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.19199999999999995
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.1
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.6466666666666667
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.7633333333333333
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.87
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.91
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.7844876010568872
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.7531666666666667
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.7413336663336665
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: 0.3
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.48
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.54
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.66
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.3
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.2
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.156
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.098
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.13885714285714287
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.28949206349206347
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.34890476190476183
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.4669047619047619
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.35173862968285535
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.41173809523809524
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.27418803367117023
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: 0.84
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.92
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.96
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.96
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.84
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.44666666666666655
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.288
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.152
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.42
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.67
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.72
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.76
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.7396469666031694
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.8846666666666667
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.6745663269195332
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: 0.4
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.56
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.64
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.72
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.4
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.18666666666666668
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.128
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.07200000000000001
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.4
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.56
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.64
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.72
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.5513379104118443
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.4978571428571428
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.5100075204701912
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: 0.42
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.5
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.58
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.62
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.42
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.36
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.316
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.256
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.040942245985757866
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.07446981664033472
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.08961908265974948
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.11750646414430962
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.31817112092123645
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.4845555555555556
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.12896096534086632
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: 0.34
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.62
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.66
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.72
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.34
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.20666666666666664
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.132
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.07400000000000001
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.32
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.58
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.62
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.68
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.5190287199365009
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.4836031746031745
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.4681591756850757
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: 0.82
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.9
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.9
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.92
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.82
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.32666666666666666
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.20799999999999996
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.11199999999999999
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.7440000000000001
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.8306666666666667
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.8540000000000001
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.8773333333333332
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.8467657844266899
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.8625
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.8312680912657447
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: 0.32
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.52
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.58
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.66
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.32
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.21333333333333332
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.18
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.12
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.066
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.13366666666666668
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.18566666666666662
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.24566666666666667
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.2462853863944484
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.4382222222222222
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.17492730667100076
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: 0.1
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.3
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.4
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.46
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.1
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.09999999999999998
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.08000000000000002
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.046000000000000006
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.1
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.3
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.4
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.46
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.275284156147708
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.21590476190476193
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.22481590517812997
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: 0.48
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.64
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.72
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.78
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.48
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.22666666666666668
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.15999999999999998
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.088
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.445
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.62
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.71
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.78
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.6190838299940942
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.5744444444444444
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.5693936722581301
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: 0.673469387755102
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.9183673469387755
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: 0.673469387755102
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.6462585034013605
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.616326530612245
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.4938775510204082
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.04572557745748646
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.12524012746937077
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.19598845564624598
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.3058088560800141
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.5544751618654643
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.7962585034013605
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.37144809611649504
name: Maxsim Map@100
- task:
type: nano-beir
name: Nano BEIR
dataset:
name: NanoBEIR mean
type: NanoBEIR_mean
metrics:
- type: MaxSim_accuracy@1
value: 0.4825745682888539
name: Maxsim Accuracy@1
- type: MaxSim_accuracy@3
value: 0.6367974882260596
name: Maxsim Accuracy@3
- type: MaxSim_accuracy@5
value: 0.7015384615384617
name: Maxsim Accuracy@5
- type: MaxSim_accuracy@10
value: 0.7584615384615384
name: Maxsim Accuracy@10
- type: MaxSim_precision@1
value: 0.4825745682888539
name: Maxsim Precision@1
- type: MaxSim_precision@3
value: 0.2979173207744636
name: Maxsim Precision@3
- type: MaxSim_precision@5
value: 0.23448665620094195
name: Maxsim Precision@5
- type: MaxSim_precision@10
value: 0.16491365777080064
name: Maxsim Precision@10
- type: MaxSim_recall@1
value: 0.2728097120697778
name: Maxsim Recall@1
- type: MaxSim_recall@3
value: 0.4050555467234125
name: Maxsim Recall@3
- type: MaxSim_recall@5
value: 0.4649483089214964
name: Maxsim Recall@5
- type: MaxSim_recall@10
value: 0.5320225280999634
name: Maxsim Recall@10
- type: MaxSim_ndcg@10
value: 0.5068836487292541
name: Maxsim Ndcg@10
- type: MaxSim_mrr@10
value: 0.573317809174952
name: Maxsim Mrr@10
- type: MaxSim_map@100
value: 0.4287074345842918
name: Maxsim Map@100
---
# ColBERT MUVERA Pico
This is a [PyLate](https://github.com/lightonai/pylate) model finetuned from [neuml/bert-hash-pico](https://huggingface.co/neuml/bert-hash-pico) on the [msmarco-en-bge-gemma unnormalized split](https://huggingface.co/datasets/lightonai/ms-marco-en-bge-gemma) dataset. It maps sentences & paragraphs to sequences of 80-dimensional dense vectors and can be used for semantic textual similarity using the MaxSim operator.
This model is trained with un-normalized scores, making it compatible with [MUVERA fixed-dimensional encoding](https://arxiv.org/abs/2405.19504).
## Usage (txtai)
This model can be used to build embeddings databases with [txtai](https://github.com/neuml/txtai) for semantic search and/or as a knowledge source for retrieval augmented generation (RAG).
_Note: txtai 9.0+ is required for late interaction model support_
```python
import txtai
embeddings = txtai.Embeddings(
path="neuml/colbert-muvera-pico",
content=True
)
embeddings.index(documents())
# Run a query
embeddings.search("query to run")
```
Late interaction models excel as reranker pipelines.
```python
from txtai.pipeline import Reranker, Similarity
similarity = Similarity(path="neuml/colbert-muvera-pico", lateencode=True)
ranker = Reranker(embeddings, similarity)
ranker("query to run")
```
## Usage (PyLate)
Alternatively, the model can be loaded with [PyLate](https://github.com/lightonai/pylate).
```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="neuml/colbert-muvera-pico",
)
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,
)
```
### Full Model Architecture
```
ColBERT(
(0): Transformer({'max_seq_length': 299, 'do_lower_case': False}) with Transformer model: BertHashModel
(1): Dense({'in_features': 80, 'out_features': 128, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity'})
)
```
## Evaluation
### BEIR Subset
The following table shows a subset of BEIR scored with the [txtai benchmarks script](https://github.com/neuml/txtai/blob/master/examples/benchmarks.py).
Scores reported are `ndcg@10` and grouped into the following three categories.
#### FULL multi-vector maxsim
| Model | Parameters | NFCorpus | SciDocs | SciFact | Average |
|:------------------|:-----------|:---------|:---------|:--------|:--------|
| [ColBERT v2](https://huggingface.co/colbert-ir/colbertv2.0) | 110M | 0.3165 | 0.1497 | 0.6456 | 0.3706 |
| [ColBERT MUVERA Femto](https://huggingface.co/neuml/colbert-muvera-femto) | 0.2M | 0.2513 | 0.0870 | 0.4710 | 0.2698 |
| [**ColBERT MUVERA Pico**](https://huggingface.co/neuml/colbert-muvera-pico) | **0.4M** | **0.3005** | **0.1117** | **0.6452** | **0.3525** |
| [ColBERT MUVERA Nano](https://huggingface.co/neuml/colbert-muvera-nano) | 0.9M | 0.3180 | 0.1262 | 0.6576 | 0.3673 |
| [ColBERT MUVERA Micro](https://huggingface.co/neuml/colbert-muvera-micro) | 4M | 0.3235 | 0.1244 | 0.6676 | 0.3718 |
#### MUVERA encoding + maxsim re-ranking of the top 100 results per MUVERA paper
| Model | Parameters | NFCorpus | SciDocs | SciFact | Average |
|:------------------|:-----------|:---------|:---------|:--------|:--------|
| [ColBERT v2](https://huggingface.co/colbert-ir/colbertv2.0) | 110M | 0.3025 | 0.1538 | 0.6278 | 0.3614 |
| [ColBERT MUVERA Femto](https://huggingface.co/neuml/colbert-muvera-femto) | 0.2M | 0.2316 | 0.0858 | 0.4641 | 0.2605 |
| [**ColBERT MUVERA Pico**](https://huggingface.co/neuml/colbert-muvera-pico) | **0.4M** | **0.2821** | **0.1004** | **0.6090** | **0.3305** |
| [ColBERT MUVERA Nano](https://huggingface.co/neuml/colbert-muvera-nano) | 0.9M | 0.2996 | 0.1201 | 0.6249 | 0.3482 |
| [ColBERT MUVERA Micro](https://huggingface.co/neuml/colbert-muvera-micro) | 4M | 0.3095 | 0.1228 | 0.6464 | 0.3596 |
#### MUVERA encoding only
| Model | Parameters | NFCorpus | SciDocs | SciFact | Average |
|:------------------|:-----------|:---------|:---------|:--------|:--------|
| [ColBERT v2](https://huggingface.co/colbert-ir/colbertv2.0) | 110M | 0.2356 | 0.1229 | 0.5002 | 0.2862 |
| [ColBERT MUVERA Femto](https://huggingface.co/neuml/colbert-muvera-femto) | 0.2M | 0.1851 | 0.0411 | 0.3518 | 0.1927 |
| [**ColBERT MUVERA Pico**](https://huggingface.co/neuml/colbert-muvera-pico) | **0.4M** | **0.1926** | **0.0564** | **0.4424** | **0.2305** |
| [ColBERT MUVERA Nano](https://huggingface.co/neuml/colbert-muvera-nano) | 0.9M | 0.2355 | 0.0807 | 0.4904 | 0.2689 |
| [ColBERT MUVERA Micro](https://huggingface.co/neuml/colbert-muvera-micro) | 4M | 0.2348 | 0.0882 | 0.4875 | 0.2702 |
_Note: The scores reported don't match scores reported in the respective papers due to different default settings in the txtai benchmark scripts._
As noted earlier, models trained with min-max score normalization don't perform well with MUVERA encoding. See this [GitHub Issue](https://github.com/lightonai/pylate/issues/142) for more.
**At 450K parameters, this model does shockingly well! It's not too far off from the baseline 4M parameter model at 1/10th the size. It's also not too far off from the original ColBERT v2 model, which has 110M parameters.**
### Nano BEIR
* Dataset: `NanoBEIR_mean`
* Evaluated with pylate.evaluation.nano_beir_evaluator.NanoBEIREvaluator
| Metric | Value |
|:--------------------|:-----------|
| MaxSim_accuracy@1 | 0.4826 |
| MaxSim_accuracy@3 | 0.6368 |
| MaxSim_accuracy@5 | 0.7015 |
| MaxSim_accuracy@10 | 0.7585 |
| MaxSim_precision@1 | 0.4826 |
| MaxSim_precision@3 | 0.2979 |
| MaxSim_precision@5 | 0.2345 |
| MaxSim_precision@10 | 0.1649 |
| MaxSim_recall@1 | 0.2728 |
| MaxSim_recall@3 | 0.4051 |
| MaxSim_recall@5 | 0.4649 |
| MaxSim_recall@10 | 0.532 |
| **MaxSim_ndcg@10** | **0.5069** |
| MaxSim_mrr@10 | 0.5733 |
| MaxSim_map@100 | 0.4287 |
## Training Details
### Training Hyperparameters
#### Non-Default Hyperparameters
- `eval_strategy`: steps
- `per_device_train_batch_size`: 32
- `learning_rate`: 0.0003
- `num_train_epochs`: 1
- `warmup_ratio`: 0.05
- `fp16`: True
#### All Hyperparameters
Click to expand
- `overwrite_output_dir`: False
- `do_predict`: False
- `eval_strategy`: steps
- `prediction_loss_only`: True
- `per_device_train_batch_size`: 32
- `per_device_eval_batch_size`: 8
- `per_gpu_train_batch_size`: None
- `per_gpu_eval_batch_size`: None
- `gradient_accumulation_steps`: 1
- `eval_accumulation_steps`: None
- `torch_empty_cache_steps`: None
- `learning_rate`: 0.0003
- `weight_decay`: 0.0
- `adam_beta1`: 0.9
- `adam_beta2`: 0.999
- `adam_epsilon`: 1e-08
- `max_grad_norm`: 1.0
- `num_train_epochs`: 1
- `max_steps`: -1
- `lr_scheduler_type`: linear
- `lr_scheduler_kwargs`: {}
- `warmup_ratio`: 0.05
- `warmup_steps`: 0
- `log_level`: passive
- `log_level_replica`: warning
- `log_on_each_node`: True
- `logging_nan_inf_filter`: True
- `save_safetensors`: True
- `save_on_each_node`: False
- `save_only_model`: False
- `restore_callback_states_from_checkpoint`: False
- `no_cuda`: False
- `use_cpu`: False
- `use_mps_device`: False
- `seed`: 42
- `data_seed`: None
- `jit_mode_eval`: False
- `bf16`: False
- `fp16`: True
- `fp16_opt_level`: O1
- `half_precision_backend`: auto
- `bf16_full_eval`: False
- `fp16_full_eval`: False
- `tf32`: None
- `local_rank`: 0
- `ddp_backend`: None
- `tpu_num_cores`: None
- `tpu_metrics_debug`: False
- `debug`: []
- `dataloader_drop_last`: False
- `dataloader_num_workers`: 0
- `dataloader_prefetch_factor`: None
- `past_index`: -1
- `disable_tqdm`: False
- `remove_unused_columns`: True
- `label_names`: None
- `load_best_model_at_end`: False
- `ignore_data_skip`: False
- `fsdp`: []
- `fsdp_min_num_params`: 0
- `fsdp_config`: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
- `fsdp_transformer_layer_cls_to_wrap`: None
- `accelerator_config`: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
- `parallelism_config`: None
- `deepspeed`: None
- `label_smoothing_factor`: 0.0
- `optim`: adamw_torch_fused
- `optim_args`: None
- `adafactor`: False
- `group_by_length`: False
- `length_column_name`: length
- `project`: huggingface
- `trackio_space_id`: trackio
- `ddp_find_unused_parameters`: None
- `ddp_bucket_cap_mb`: None
- `ddp_broadcast_buffers`: False
- `dataloader_pin_memory`: True
- `dataloader_persistent_workers`: False
- `skip_memory_metrics`: True
- `use_legacy_prediction_loop`: False
- `push_to_hub`: False
- `resume_from_checkpoint`: None
- `hub_model_id`: None
- `hub_strategy`: every_save
- `hub_private_repo`: None
- `hub_always_push`: False
- `hub_revision`: None
- `gradient_checkpointing`: False
- `gradient_checkpointing_kwargs`: None
- `include_inputs_for_metrics`: False
- `include_for_metrics`: []
- `eval_do_concat_batches`: True
- `fp16_backend`: auto
- `push_to_hub_model_id`: None
- `push_to_hub_organization`: None
- `mp_parameters`:
- `auto_find_batch_size`: False
- `full_determinism`: False
- `torchdynamo`: None
- `ray_scope`: last
- `ddp_timeout`: 1800
- `torch_compile`: False
- `torch_compile_backend`: None
- `torch_compile_mode`: None
- `include_tokens_per_second`: False
- `include_num_input_tokens_seen`: no
- `neftune_noise_alpha`: None
- `optim_target_modules`: None
- `batch_eval_metrics`: False
- `eval_on_start`: False
- `use_liger_kernel`: False
- `liger_kernel_config`: None
- `eval_use_gather_object`: False
- `average_tokens_across_devices`: True
- `prompts`: None
- `batch_sampler`: batch_sampler
- `multi_dataset_batch_sampler`: proportional