metadata

language:
  - en
license: apache-2.0
tags:
  - sentence-transformers
  - sparse-encoder
  - sparse
  - csr
  - generated_from_trainer
  - dataset_size:99000
  - loss:CSRLoss
  - loss:SparseMultipleNegativesRankingLoss
base_model: mixedbread-ai/mxbai-embed-large-v1
widget:
  - text: >-
      Saudi Arabia–United Arab Emirates relations However, the UAE and Saudi
      Arabia continue to take somewhat differing stances on regional conflicts
      such the Yemeni Civil War, where the UAE opposes Al-Islah, and supports
      the Southern Movement, which has fought against Saudi-backed forces, and
      the Syrian Civil War, where the UAE has disagreed with Saudi support for
      Islamist movements.[4]
  - text: >-
      Economy of New Zealand New Zealand's diverse market economy has a sizable
      service sector, accounting for 63% of all GDP activity in 2013.[17] Large
      scale manufacturing industries include aluminium production, food
      processing, metal fabrication, wood and paper products. Mining,
      manufacturing, electricity, gas, water, and waste services accounted for
      16.5% of GDP in 2013.[17] The primary sector continues to dominate New
      Zealand's exports, despite accounting for 6.5% of GDP in 2013.[17]
  - text: >-
      who was the first president of indian science congress meeting held in
      kolkata in 1914
  - text: >-
      Get Over It (Eagles song) "Get Over It" is a song by the Eagles released
      as a single after a fourteen-year breakup. It was also the first song
      written by bandmates Don Henley and Glenn Frey when the band reunited.
      "Get Over It" was played live for the first time during their Hell Freezes
      Over tour in 1994. It returned the band to the U.S. Top 40 after a
      fourteen-year absence, peaking at No. 31 on the Billboard Hot 100 chart.
      It also hit No. 4 on the Billboard Mainstream Rock Tracks chart. The song
      was not played live by the Eagles after the "Hell Freezes Over" tour in
      1994. It remains the group's last Top 40 hit in the U.S.
  - text: >-
      Cornelius the Centurion Cornelius (Greek: Κορνήλιος) was a Roman centurion
      who is considered by Christians to be one of the first Gentiles to convert
      to the faith, as related in Acts of the Apostles.
datasets:
  - sentence-transformers/natural-questions
pipeline_tag: feature-extraction
library_name: sentence-transformers
metrics:
  - cosine_accuracy@1
  - cosine_accuracy@3
  - cosine_accuracy@5
  - cosine_accuracy@10
  - cosine_precision@1
  - cosine_precision@3
  - cosine_precision@5
  - cosine_precision@10
  - cosine_recall@1
  - cosine_recall@3
  - cosine_recall@5
  - cosine_recall@10
  - cosine_ndcg@10
  - cosine_mrr@10
  - cosine_map@100
  - query_active_dims
  - query_sparsity_ratio
  - corpus_active_dims
  - corpus_sparsity_ratio
co2_eq_emissions:
  emissions: 39.03404179469692
  energy_consumed: 0.1004215100377588
  source: codecarbon
  training_type: fine-tuning
  on_cloud: false
  cpu_model: 13th Gen Intel(R) Core(TM) i7-13700K
  ram_total_size: 31.777088165283203
  hours_used: 0.246
  hardware_used: 1 x NVIDIA GeForce RTX 3090
model-index:
  - name: Sparse CSR model trained on Natural Questions
    results:
      - task:
          type: sparse-information-retrieval
          name: Sparse Information Retrieval
        dataset:
          name: nq eval 4
          type: nq_eval_4
        metrics:
          - type: cosine_accuracy@1
            value: 0.333
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.51
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.608
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.701
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.333
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.17
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.1216
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.0701
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.333
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.51
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.608
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.701
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.5048911324016669
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.44326626984126943
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.45271073834573333
            name: Cosine Map@100
          - type: query_active_dims
            value: 4
            name: Query Active Dims
          - type: query_sparsity_ratio
            value: 0.9990234375
            name: Query Sparsity Ratio
          - type: corpus_active_dims
            value: 4
            name: Corpus Active Dims
          - type: corpus_sparsity_ratio
            value: 0.9990234375
            name: Corpus Sparsity Ratio
      - task:
          type: sparse-information-retrieval
          name: Sparse Information Retrieval
        dataset:
          name: nq eval 8
          type: nq_eval_8
        metrics:
          - type: cosine_accuracy@1
            value: 0.471
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.675
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.75
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.825
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.471
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.225
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.15
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.0825
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.471
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.675
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.75
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.825
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.6441336669789526
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.5863865079365083
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.5935240561774322
            name: Cosine Map@100
          - type: query_active_dims
            value: 8
            name: Query Active Dims
          - type: query_sparsity_ratio
            value: 0.998046875
            name: Query Sparsity Ratio
          - type: corpus_active_dims
            value: 8
            name: Corpus Active Dims
          - type: corpus_sparsity_ratio
            value: 0.998046875
            name: Corpus Sparsity Ratio
      - task:
          type: sparse-information-retrieval
          name: Sparse Information Retrieval
        dataset:
          name: nq eval 16
          type: nq_eval_16
        metrics:
          - type: cosine_accuracy@1
            value: 0.618
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.8
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.839
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.888
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.618
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.2666666666666666
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.1678
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.08880000000000002
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.618
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.8
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.839
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.888
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.7584976627273415
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7165746031746036
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7213877485505877
            name: Cosine Map@100
          - type: query_active_dims
            value: 16
            name: Query Active Dims
          - type: query_sparsity_ratio
            value: 0.99609375
            name: Query Sparsity Ratio
          - type: corpus_active_dims
            value: 16
            name: Corpus Active Dims
          - type: corpus_sparsity_ratio
            value: 0.99609375
            name: Corpus Sparsity Ratio
      - task:
          type: sparse-information-retrieval
          name: Sparse Information Retrieval
        dataset:
          name: nq eval 32
          type: nq_eval_32
        metrics:
          - type: cosine_accuracy@1
            value: 0.729
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.848
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.881
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.916
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.729
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.2826666666666666
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.1762
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09160000000000001
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.729
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.848
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.881
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.916
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.8242272725827696
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7946277777777779
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7980770968534903
            name: Cosine Map@100
          - type: query_active_dims
            value: 32
            name: Query Active Dims
          - type: query_sparsity_ratio
            value: 0.9921875
            name: Query Sparsity Ratio
          - type: corpus_active_dims
            value: 32
            name: Corpus Active Dims
          - type: corpus_sparsity_ratio
            value: 0.9921875
            name: Corpus Sparsity Ratio
      - task:
          type: sparse-information-retrieval
          name: Sparse Information Retrieval
        dataset:
          name: nq eval 64
          type: nq_eval_64
        metrics:
          - type: cosine_accuracy@1
            value: 0.783
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.883
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.909
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.94
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.783
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.29433333333333334
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.18180000000000002
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09400000000000001
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.783
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.883
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.909
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.94
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.8633645356650496
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.8386107142857145
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.8412127714611879
            name: Cosine Map@100
          - type: query_active_dims
            value: 64
            name: Query Active Dims
          - type: query_sparsity_ratio
            value: 0.984375
            name: Query Sparsity Ratio
          - type: corpus_active_dims
            value: 64
            name: Corpus Active Dims
          - type: corpus_sparsity_ratio
            value: 0.984375
            name: Corpus Sparsity Ratio
      - task:
          type: sparse-information-retrieval
          name: Sparse Information Retrieval
        dataset:
          name: nq eval 128
          type: nq_eval_128
        metrics:
          - type: cosine_accuracy@1
            value: 0.858
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.942
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.953
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.966
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.858
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.31399999999999995
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.19060000000000005
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.0966
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.858
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.942
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.953
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.966
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.9175695694881496
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.9015206349206352
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.9028827893432363
            name: Cosine Map@100
          - type: query_active_dims
            value: 128
            name: Query Active Dims
          - type: query_sparsity_ratio
            value: 0.96875
            name: Query Sparsity Ratio
          - type: corpus_active_dims
            value: 128
            name: Corpus Active Dims
          - type: corpus_sparsity_ratio
            value: 0.96875
            name: Corpus Sparsity Ratio
      - task:
          type: sparse-information-retrieval
          name: Sparse Information Retrieval
        dataset:
          name: nq eval 256
          type: nq_eval_256
        metrics:
          - type: cosine_accuracy@1
            value: 0.905
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.972
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.982
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.987
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.905
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.32399999999999995
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.19640000000000005
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09870000000000002
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.905
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.972
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.982
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.987
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.9511220239850359
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.9390623015873019
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.9396249937318298
            name: Cosine Map@100
          - type: query_active_dims
            value: 256
            name: Query Active Dims
          - type: query_sparsity_ratio
            value: 0.9375
            name: Query Sparsity Ratio
          - type: corpus_active_dims
            value: 256
            name: Corpus Active Dims
          - type: corpus_sparsity_ratio
            value: 0.9375
            name: Corpus Sparsity Ratio

Sparse CSR model trained on Natural Questions

This is a CSR Sparse Encoder model finetuned from mixedbread-ai/mxbai-embed-large-v1 on the natural-questions dataset using the sentence-transformers library. It maps sentences & paragraphs to a 4096-dimensional sparse vector space with 256 maximum active dimensions and can be used for semantic search and sparse retrieval.

Model Details

Model Description

Model Type: CSR Sparse Encoder
Base model: mixedbread-ai/mxbai-embed-large-v1
Maximum Sequence Length: 512 tokens
Output Dimensionality: 4096 dimensions (trained with 256 maximum active dimensions)
Similarity Function: Cosine Similarity
Training Dataset:
- natural-questions
Language: en
License: apache-2.0

Model Sources

Documentation: Sentence Transformers Documentation
Documentation: Sparse Encoder Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sparse Encoders on Hugging Face

Full Model Architecture

SparseEncoder(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): CSRSparsity({'input_dim': 1024, 'hidden_dim': 4096, 'k': 256, 'k_aux': 512, 'normalize': False, 'dead_threshold': 30})
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SparseEncoder

# Download from the 🤗 Hub
model = SparseEncoder("tomaarsen/csr-mxbai-embed-large-v1-nq-cos-sim-scale-5-gamma-0.1-detach-2")
# Run inference
queries = [
    "who is cornelius in the book of acts",
]
documents = [
    'Cornelius the Centurion Cornelius (Greek: Κορνήλιος) was a Roman centurion who is considered by Christians to be one of the first Gentiles to convert to the faith, as related in Acts of the Apostles.',
    "Joe Ranft Ranft reunited with Lasseter when he was hired by Pixar in 1991 as their head of story.[1] There he worked on all of their films produced up to 2006; this included Toy Story (for which he received an Academy Award nomination) and A Bug's Life, as the co-story writer and others as story supervisor. His final film was Cars. He also voiced characters in many of the films, including Heimlich the caterpillar in A Bug's Life, Wheezy the penguin in Toy Story 2, and Jacques the shrimp in Finding Nemo.[1]",
    'Wonderful Tonight "Wonderful Tonight" is a ballad written by Eric Clapton. It was included on Clapton\'s 1977 album Slowhand. Clapton wrote the song about Pattie Boyd.[1] The female vocal harmonies on the song are provided by Marcella Detroit (then Marcy Levy) and Yvonne Elliman.',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 4096] [3, 4096]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[0.7512, 0.2310, 0.2134]])

Evaluation

Metrics

Sparse Information Retrieval

Dataset: nq_eval_4
Evaluated with SparseInformationRetrievalEvaluator with these parameters:
```
{
    "max_active_dims": 4
}
```

Metric	Value
cosine_accuracy@1	0.333
cosine_accuracy@3	0.51
cosine_accuracy@5	0.608
cosine_accuracy@10	0.701
cosine_precision@1	0.333
cosine_precision@3	0.17
cosine_precision@5	0.1216
cosine_precision@10	0.0701
cosine_recall@1	0.333
cosine_recall@3	0.51
cosine_recall@5	0.608
cosine_recall@10	0.701
cosine_ndcg@10	0.5049
cosine_mrr@10	0.4433
cosine_map@100	0.4527
query_active_dims	4.0
query_sparsity_ratio	0.999
corpus_active_dims	4.0
corpus_sparsity_ratio	0.999

Sparse Information Retrieval

Dataset: nq_eval_8
Evaluated with SparseInformationRetrievalEvaluator with these parameters:
```
{
    "max_active_dims": 8
}
```

Metric	Value
cosine_accuracy@1	0.471
cosine_accuracy@3	0.675
cosine_accuracy@5	0.75
cosine_accuracy@10	0.825
cosine_precision@1	0.471
cosine_precision@3	0.225
cosine_precision@5	0.15
cosine_precision@10	0.0825
cosine_recall@1	0.471
cosine_recall@3	0.675
cosine_recall@5	0.75
cosine_recall@10	0.825
cosine_ndcg@10	0.6441
cosine_mrr@10	0.5864
cosine_map@100	0.5935
query_active_dims	8.0
query_sparsity_ratio	0.998
corpus_active_dims	8.0
corpus_sparsity_ratio	0.998

Sparse Information Retrieval

Dataset: nq_eval_16
Evaluated with SparseInformationRetrievalEvaluator with these parameters:
```
{
    "max_active_dims": 16
}
```

Metric	Value
cosine_accuracy@1	0.618
cosine_accuracy@3	0.8
cosine_accuracy@5	0.839
cosine_accuracy@10	0.888
cosine_precision@1	0.618
cosine_precision@3	0.2667
cosine_precision@5	0.1678
cosine_precision@10	0.0888
cosine_recall@1	0.618
cosine_recall@3	0.8
cosine_recall@5	0.839
cosine_recall@10	0.888
cosine_ndcg@10	0.7585
cosine_mrr@10	0.7166
cosine_map@100	0.7214
query_active_dims	16.0
query_sparsity_ratio	0.9961
corpus_active_dims	16.0
corpus_sparsity_ratio	0.9961

Sparse Information Retrieval

Dataset: nq_eval_32
Evaluated with SparseInformationRetrievalEvaluator with these parameters:
```
{
    "max_active_dims": 32
}
```

Metric	Value
cosine_accuracy@1	0.729
cosine_accuracy@3	0.848
cosine_accuracy@5	0.881
cosine_accuracy@10	0.916
cosine_precision@1	0.729
cosine_precision@3	0.2827
cosine_precision@5	0.1762
cosine_precision@10	0.0916
cosine_recall@1	0.729
cosine_recall@3	0.848
cosine_recall@5	0.881
cosine_recall@10	0.916
cosine_ndcg@10	0.8242
cosine_mrr@10	0.7946
cosine_map@100	0.7981
query_active_dims	32.0
query_sparsity_ratio	0.9922
corpus_active_dims	32.0
corpus_sparsity_ratio	0.9922

Sparse Information Retrieval

Dataset: nq_eval_64
Evaluated with SparseInformationRetrievalEvaluator with these parameters:
```
{
    "max_active_dims": 64
}
```

Metric	Value
cosine_accuracy@1	0.783
cosine_accuracy@3	0.883
cosine_accuracy@5	0.909
cosine_accuracy@10	0.94
cosine_precision@1	0.783
cosine_precision@3	0.2943
cosine_precision@5	0.1818
cosine_precision@10	0.094
cosine_recall@1	0.783
cosine_recall@3	0.883
cosine_recall@5	0.909
cosine_recall@10	0.94
cosine_ndcg@10	0.8634
cosine_mrr@10	0.8386
cosine_map@100	0.8412
query_active_dims	64.0
query_sparsity_ratio	0.9844
corpus_active_dims	64.0
corpus_sparsity_ratio	0.9844

Sparse Information Retrieval

Dataset: nq_eval_128
Evaluated with SparseInformationRetrievalEvaluator with these parameters:
```
{
    "max_active_dims": 128
}
```

Metric	Value
cosine_accuracy@1	0.858
cosine_accuracy@3	0.942
cosine_accuracy@5	0.953
cosine_accuracy@10	0.966
cosine_precision@1	0.858
cosine_precision@3	0.314
cosine_precision@5	0.1906
cosine_precision@10	0.0966
cosine_recall@1	0.858
cosine_recall@3	0.942
cosine_recall@5	0.953
cosine_recall@10	0.966
cosine_ndcg@10	0.9176
cosine_mrr@10	0.9015
cosine_map@100	0.9029
query_active_dims	128.0
query_sparsity_ratio	0.9688
corpus_active_dims	128.0
corpus_sparsity_ratio	0.9688

Sparse Information Retrieval

Dataset: nq_eval_256
Evaluated with SparseInformationRetrievalEvaluator with these parameters:
```
{
    "max_active_dims": 256
}
```

Metric	Value
cosine_accuracy@1	0.905
cosine_accuracy@3	0.972
cosine_accuracy@5	0.982
cosine_accuracy@10	0.987
cosine_precision@1	0.905
cosine_precision@3	0.324
cosine_precision@5	0.1964
cosine_precision@10	0.0987
cosine_recall@1	0.905
cosine_recall@3	0.972
cosine_recall@5	0.982
cosine_recall@10	0.987
cosine_ndcg@10	0.9511
cosine_mrr@10	0.9391
cosine_map@100	0.9396
query_active_dims	256.0
query_sparsity_ratio	0.9375
corpus_active_dims	256.0
corpus_sparsity_ratio	0.9375

Training Details

Training Dataset

natural-questions

Dataset: natural-questions at f9e894e
Size: 99,000 training samples
Columns: query and answer
Approximate statistics based on the first 1000 samples:
query answer
type string string
details
min: 10 tokens
mean: 11.71 tokens
max: 26 tokens

min: 4 tokens
mean: 131.81 tokens
max: 450 tokens

	query	answer
type	string	string
details	min: 10 tokens mean: 11.71 tokens max: 26 tokens	min: 4 tokens mean: 131.81 tokens max: 450 tokens

Samples:

query	answer
`who played the father in papa don't preach`	`Alex McArthur Alex McArthur (born March 6, 1957) is an American actor.`
`where was the location of the battle of hastings`	`Battle of Hastings The Battle of Hastings[a] was fought on 14 October 1066 between the Norman-French army of William, the Duke of Normandy, and an English army under the Anglo-Saxon King Harold Godwinson, beginning the Norman conquest of England. It took place approximately 7 miles (11 kilometres) northwest of Hastings, close to the present-day town of Battle, East Sussex, and was a decisive Norman victory.`
`how many puppies can a dog give birth to`	`Canine reproduction The largest litter size to date was set by a Neapolitan Mastiff in Manea, Cambridgeshire, UK on November 29, 2004; the litter was 24 puppies.[22]`

Loss: CSRLoss with these parameters:

{
    "beta": 0.1,
    "gamma": 0.1,
    "loss": "SparseMultipleNegativesRankingLoss(scale=5.0, similarity_fct='cos_sim')"
}

Evaluation Dataset

natural-questions

Dataset: natural-questions at f9e894e
Size: 1,000 evaluation samples
Columns: query and answer
Approximate statistics based on the first 1000 samples:
query answer
type string string
details
min: 10 tokens
mean: 11.69 tokens
max: 23 tokens

min: 15 tokens
mean: 134.01 tokens
max: 512 tokens

	query	answer
type	string	string
details	min: 10 tokens mean: 11.69 tokens max: 23 tokens	min: 15 tokens mean: 134.01 tokens max: 512 tokens

Samples:

query	answer
`where is the tiber river located in italy`	Tiber The Tiber (/ˈtaɪbər/, Latin: Tiberis,[1] Italian: Tevere [ˈteːvere])[2] is the third-longest river in Italy, rising in the Apennine Mountains in Emilia-Romagna and flowing 406 kilometres (252 mi) through Tuscany, Umbria and Lazio, where it is joined by the river Aniene, to the Tyrrhenian Sea, between Ostia and Fiumicino.[3] It drains a basin estimated at 17,375 square kilometres (6,709 sq mi). The river has achieved lasting fame as the main watercourse of the city of Rome, founded on its eastern banks.
`what kind of car does jay gatsby drive`	`Jay Gatsby At the Buchanan home, Jordan Baker, Nick, Jay, and the Buchanans decide to visit New York City. Tom borrows Gatsby's yellow Rolls Royce to drive up to the city. On the way to New York City, Tom makes a detour at a gas station in "the Valley of Ashes", a run-down part of Long Island. The owner, George Wilson, shares his concern that his wife, Myrtle, may be having an affair. This unnerves Tom, who has been having an affair with Myrtle, and he leaves in a hurry.`
`who sings if i can dream about you`	`I Can Dream About You "I Can Dream About You" is a song performed by American singer Dan Hartman on the soundtrack album of the film Streets of Fire. Released in 1984 as a single from the soundtrack, and included on Hartman's album I Can Dream About You, it reached number 6 on the Billboard Hot 100.[1]`

Loss: CSRLoss with these parameters:

{
    "beta": 0.1,
    "gamma": 0.1,
    "loss": "SparseMultipleNegativesRankingLoss(scale=5.0, similarity_fct='cos_sim')"
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 64
per_device_eval_batch_size: 64
learning_rate: 4e-05
num_train_epochs: 1
bf16: True
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 64
per_device_eval_batch_size: 64
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: 4e-05
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: 0.0
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
use_ipex: False
bf16: True
fp16: False
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}
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_torch
optim_args: None
adafactor: False
group_by_length: False
length_column_name: length
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
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: False
neftune_noise_alpha: None
optim_target_modules: None
batch_eval_metrics: False
eval_on_start: False
use_liger_kernel: False
eval_use_gather_object: False
average_tokens_across_devices: False
prompts: None
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss	nq_eval_4_cosine_ndcg@10	nq_eval_8_cosine_ndcg@10	nq_eval_16_cosine_ndcg@10	nq_eval_32_cosine_ndcg@10	nq_eval_64_cosine_ndcg@10	nq_eval_128_cosine_ndcg@10	nq_eval_256_cosine_ndcg@10
-1	-1	-	-	0.2820	0.4878	0.6965	0.8627	0.9319	0.9578	0.9699
0.0646	100	0.5473	-	-	-	-	-	-	-	-
0.1293	200	0.4992	-	-	-	-	-	-	-	-
0.1939	300	0.4823	0.4529	0.4274	0.6028	0.7463	0.8377	0.8877	0.9224	0.9512
0.2586	400	0.4725	-	-	-	-	-	-	-	-
0.3232	500	0.4655	-	-	-	-	-	-	-	-
0.3878	600	0.4597	0.4344	0.4642	0.6281	0.7556	0.8281	0.8697	0.9163	0.9485
0.4525	700	0.4563	-	-	-	-	-	-	-	-
0.5171	800	0.4522	-	-	-	-	-	-	-	-
0.5818	900	0.4496	0.4256	0.4908	0.6406	0.7527	0.8232	0.8653	0.9146	0.9467
0.6464	1000	0.4478	-	-	-	-	-	-	-	-
0.7111	1100	0.4458	-	-	-	-	-	-	-	-
0.7757	1200	0.4448	0.4210	0.5008	0.6424	0.7659	0.8186	0.8649	0.9151	0.9502
0.8403	1300	0.4436	-	-	-	-	-	-	-	-
0.9050	1400	0.4425	-	-	-	-	-	-	-	-
0.9696	1500	0.4427	0.4193	0.5064	0.6434	0.7584	0.8229	0.8646	0.9178	0.9517
-1	-1	-	-	0.5049	0.6441	0.7585	0.8242	0.8634	0.9176	0.9511

Environmental Impact

Carbon emissions were measured using CodeCarbon.

Energy Consumed: 0.100 kWh
Carbon Emitted: 0.039 kg of CO2
Hours Used: 0.246 hours

Training Hardware

On Cloud: No
GPU Model: 1 x NVIDIA GeForce RTX 3090
CPU Model: 13th Gen Intel(R) Core(TM) i7-13700K
RAM Size: 31.78 GB

Framework Versions

Python: 3.11.6
Sentence Transformers: 4.2.0.dev0
Transformers: 4.52.4
PyTorch: 2.6.0+cu124
Accelerate: 1.5.1
Datasets: 2.21.0
Tokenizers: 0.21.1

Citation

BibTeX

Sentence Transformers

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}

CSRLoss

@misc{wen2025matryoshkarevisitingsparsecoding,
      title={Beyond Matryoshka: Revisiting Sparse Coding for Adaptive Representation},
      author={Tiansheng Wen and Yifei Wang and Zequn Zeng and Zhong Peng and Yudi Su and Xinyang Liu and Bo Chen and Hongwei Liu and Stefanie Jegelka and Chenyu You},
      year={2025},
      eprint={2503.01776},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2503.01776},
}

SparseMultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}