tomaarsen/csr-mxbai-embed-large-v1-nq

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 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
• Similarity Function: Dot Product
• 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")
# Run inference
sentences = [
    'who is cornelius in the book of acts',
    '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]",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# (3, 4096)

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Sparse Information Retrieval

• Datasets: NanoMSMARCO, NanoNFCorpus, NanoNQ, NanoMSMARCO, NanoNFCorpus, NanoNQ, NanoMSMARCO, NanoNFCorpus, NanoNQ, NanoMSMARCO, NanoNFCorpus, NanoNQ, NanoClimateFEVER, NanoDBPedia, NanoFEVER, NanoFiQA2018, NanoHotpotQA, NanoMSMARCO, NanoNFCorpus, NanoNQ, NanoQuoraRetrieval, NanoSCIDOCS, NanoArguAna, NanoSciFact and NanoTouche2020
• Evaluated with SparseInformationRetrievalEvaluator

Metric	NanoMSMARCO	NanoNFCorpus	NanoNQ	NanoClimateFEVER	NanoDBPedia	NanoFEVER	NanoFiQA2018	NanoHotpotQA	NanoQuoraRetrieval	NanoSCIDOCS	NanoArguAna	NanoSciFact	NanoTouche2020
dot_accuracy@1	0.4	0.42	0.48	0.28	0.74	0.84	0.4	0.78	0.92	0.46	0.32	0.7	0.5306
dot_accuracy@3	0.68	0.56	0.72	0.48	0.9	0.96	0.64	0.94	0.98	0.66	0.82	0.72	0.8571
dot_accuracy@5	0.76	0.6	0.76	0.56	0.92	0.96	0.7	0.98	1.0	0.74	0.88	0.78	0.898
dot_accuracy@10	0.82	0.68	0.84	0.64	0.98	0.96	0.78	1.0	1.0	0.86	0.96	0.86	0.9592
dot_precision@1	0.4	0.42	0.48	0.28	0.74	0.84	0.4	0.78	0.92	0.46	0.32	0.7	0.5306
dot_precision@3	0.2267	0.36	0.2467	0.18	0.64	0.3267	0.2933	0.5267	0.4067	0.34	0.2733	0.2667	0.551
dot_precision@5	0.152	0.32	0.156	0.136	0.592	0.2	0.224	0.336	0.26	0.28	0.176	0.172	0.498
dot_precision@10	0.082	0.27	0.09	0.086	0.468	0.102	0.136	0.18	0.136	0.198	0.096	0.096	0.4163
dot_recall@1	0.4	0.0464	0.47	0.115	0.0898	0.7867	0.2072	0.39	0.8073	0.0977	0.32	0.665	0.0391
dot_recall@3	0.68	0.0776	0.68	0.2117	0.1711	0.9167	0.4125	0.79	0.942	0.2137	0.82	0.715	0.1155
dot_recall@5	0.76	0.095	0.71	0.2757	0.2382	0.9233	0.5159	0.84	0.976	0.2897	0.88	0.765	0.1737
dot_recall@10	0.82	0.1265	0.8	0.334	0.3593	0.9333	0.6268	0.9	0.9933	0.4057	0.96	0.85	0.2788
dot_ndcg@10	0.6233	0.3262	0.6448	0.2809	0.6049	0.8812	0.488	0.8241	0.9567	0.3897	0.6618	0.7556	0.4666
dot_mrr@10	0.559	0.5004	0.6067	0.3961	0.8312	0.89	0.5356	0.8729	0.955	0.5808	0.563	0.7335	0.6902
dot_map@100	0.5667	0.1527	0.5961	0.2205	0.4433	0.8538	0.4061	0.7644	0.9393	0.3103	0.5655	0.7269	0.3537
row_non_zero_mean_query	256.0	256.0	256.0	256.0	256.0	256.0	256.0	256.0	256.0	256.0	256.0	256.0	256.0
row_sparsity_mean_query	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375
row_non_zero_mean_corpus	256.0	256.0	256.0	256.0	256.0	256.0	256.0	256.0	256.0	256.0	256.0	256.0	256.0
row_sparsity_mean_corpus	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375	0.9375

Sparse Nano BEIR

• Dataset: NanoBEIR_mean
• Evaluated with SparseNanoBEIREvaluator with these parameters:

  {
      "dataset_names": [
          "msmarco",
          "nfcorpus",
          "nq"
      ]
  }
  

Metric	Value
dot_accuracy@1	0.28
dot_accuracy@3	0.4133
dot_accuracy@5	0.52
dot_accuracy@10	0.6733
dot_precision@1	0.28
dot_precision@3	0.16
dot_precision@5	0.132
dot_precision@10	0.1007
dot_recall@1	0.2085
dot_recall@3	0.3087
dot_recall@5	0.4064
dot_recall@10	0.5136
dot_ndcg@10	0.3888
dot_mrr@10	0.383
dot_map@100	0.3054
row_non_zero_mean_query	32.0
row_sparsity_mean_query	0.9922
row_non_zero_mean_corpus	32.0
row_sparsity_mean_corpus	0.9922

Sparse Nano BEIR

• Dataset: NanoBEIR_mean
• Evaluated with SparseNanoBEIREvaluator with these parameters:

  {
      "dataset_names": [
          "msmarco",
          "nfcorpus",
          "nq"
      ]
  }
  

Metric	Value
dot_accuracy@1	0.3267
dot_accuracy@3	0.5533
dot_accuracy@5	0.6333
dot_accuracy@10	0.7133
dot_precision@1	0.3267
dot_precision@3	0.2178
dot_precision@5	0.168
dot_precision@10	0.1133
dot_recall@1	0.2434
dot_recall@3	0.4251
dot_recall@5	0.4838
dot_recall@10	0.5447
dot_ndcg@10	0.4526
dot_mrr@10	0.457
dot_map@100	0.3705
row_non_zero_mean_query	64.0
row_sparsity_mean_query	0.9844
row_non_zero_mean_corpus	64.0
row_sparsity_mean_corpus	0.9844

Sparse Nano BEIR

• Dataset: NanoBEIR_mean
• Evaluated with SparseNanoBEIREvaluator with these parameters:

  {
      "dataset_names": [
          "msmarco",
          "nfcorpus",
          "nq"
      ]
  }
  

Metric	Value
dot_accuracy@1	0.4
dot_accuracy@3	0.6333
dot_accuracy@5	0.68
dot_accuracy@10	0.72
dot_precision@1	0.4
dot_precision@3	0.2622
dot_precision@5	0.1947
dot_precision@10	0.1307
dot_recall@1	0.2937
dot_recall@3	0.4775
dot_recall@5	0.518
dot_recall@10	0.5473
dot_ndcg@10	0.5036
dot_mrr@10	0.5229
dot_map@100	0.4213
row_non_zero_mean_query	128.0
row_sparsity_mean_query	0.9688
row_non_zero_mean_corpus	128.0
row_sparsity_mean_corpus	0.9688

Sparse Nano BEIR

• Dataset: NanoBEIR_mean
• Evaluated with SparseNanoBEIREvaluator with these parameters:

  {
      "dataset_names": [
          "msmarco",
          "nfcorpus",
          "nq"
      ]
  }
  

Metric	Value
dot_accuracy@1	0.3867
dot_accuracy@3	0.6067
dot_accuracy@5	0.7
dot_accuracy@10	0.8133
dot_precision@1	0.3867
dot_precision@3	0.2511
dot_precision@5	0.2067
dot_precision@10	0.154
dot_recall@1	0.2867
dot_recall@3	0.4548
dot_recall@5	0.5026
dot_recall@10	0.5979
dot_ndcg@10	0.5238
dot_mrr@10	0.5265
dot_map@100	0.4211
row_non_zero_mean_query	256.0
row_sparsity_mean_query	0.9375
row_non_zero_mean_corpus	256.0
row_sparsity_mean_corpus	0.9375

Sparse Nano BEIR

• Dataset: NanoBEIR_mean
• Evaluated with SparseNanoBEIREvaluator with these parameters:

  {
      "dataset_names": [
          "climatefever",
          "dbpedia",
          "fever",
          "fiqa2018",
          "hotpotqa",
          "msmarco",
          "nfcorpus",
          "nq",
          "quoraretrieval",
          "scidocs",
          "arguana",
          "scifact",
          "touche2020"
      ]
  }
  

Metric	Value
dot_accuracy@1	0.5593
dot_accuracy@3	0.7629
dot_accuracy@5	0.8106
dot_accuracy@10	0.8722
dot_precision@1	0.5593
dot_precision@3	0.3567
dot_precision@5	0.2694
dot_precision@10	0.1813
dot_recall@1	0.3411
dot_recall@3	0.5189
dot_recall@5	0.5725
dot_recall@10	0.6452
dot_ndcg@10	0.608
dot_mrr@10	0.6703
dot_map@100	0.5307
row_non_zero_mean_query	256.0
row_sparsity_mean_query	0.9375
row_non_zero_mean_corpus	256.0
row_sparsity_mean_corpus	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

• 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": 1.0,
      "loss": "SparseMultipleNegativesRankingLoss(scale=1.0, similarity_fct='dot_score')"
  }
  

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

• 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": 1.0,
      "loss": "SparseMultipleNegativesRankingLoss(scale=1.0, similarity_fct='dot_score')"
  }
  

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
• load_best_model_at_end: 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: True
• 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
• dispatch_batches: None
• split_batches: 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

Training Logs

Epoch	Step	Training Loss	Validation Loss	NanoMSMARCO_dot_ndcg@10	NanoNFCorpus_dot_ndcg@10	NanoNQ_dot_ndcg@10	NanoBEIR_mean_dot_ndcg@10	NanoClimateFEVER_dot_ndcg@10	NanoDBPedia_dot_ndcg@10	NanoFEVER_dot_ndcg@10	NanoFiQA2018_dot_ndcg@10	NanoHotpotQA_dot_ndcg@10	NanoQuoraRetrieval_dot_ndcg@10	NanoSCIDOCS_dot_ndcg@10	NanoArguAna_dot_ndcg@10	NanoSciFact_dot_ndcg@10	NanoTouche2020_dot_ndcg@10
0.0646	100	0.3429	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.1293	200	0.3521	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.1939	300	0.3399	0.3572	0.6207	0.3281	0.6434	0.5308	-	-	-	-	-	-	-	-	-	-
0.2586	400	0.3458	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.3232	500	0.3383	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.3878	600	0.3613	0.3705	0.5998	0.3108	0.6044	0.5050	-	-	-	-	-	-	-	-	-	-
0.4525	700	0.3323	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.5171	800	0.316	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.5818	900	0.3336	0.3499	0.5970	0.3092	0.6616	0.5226	-	-	-	-	-	-	-	-	-	-
0.6464	1000	0.3161	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.7111	1100	0.3329	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.7757	1200	0.3615	0.3609	0.6036	0.3108	0.6372	0.5172	-	-	-	-	-	-	-	-	-	-
0.8403	1300	0.337	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.9050	1400	0.3265	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.9696	1500	0.3246	0.3527	0.6202	0.3109	0.6404	0.5238	-	-	-	-	-	-	-	-	-	-
-1	-1	-	-	0.6233	0.3262	0.6448	0.6080	0.2809	0.6049	0.8812	0.4880	0.8241	0.9567	0.3897	0.6618	0.7556	0.4666

• The bold row denotes the saved checkpoint.

Environmental Impact

Carbon emissions were measured using CodeCarbon.

• Energy Consumed: 0.202 kWh
• Carbon Emitted: 0.079 kg of CO2
• Hours Used: 0.571 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: 4.49.0
• 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}
}