tomaarsen/csr-mxbai-embed-large-v1-gooaq-1e-5-512bs

Sparse CSR model trained on Natural Questions

This is a CSR Sparse Encoder model finetuned from mixedbread-ai/mxbai-embed-large-v1 on the gooaq 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:
  • gooaq
• 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-gooaq-1e-5-512bs")
# Run inference
sentences = [
    'are you human korean novela?',
    "Are You Human? (Korean: 너도 인간이니; RR: Neodo Inganini; lit. Are You Human Too?) is a 2018 South Korean television series starring Seo Kang-jun and Gong Seung-yeon. It aired on KBS2's Mondays and Tuesdays at 22:00 (KST) time slot, from June 4 to August 7, 2018.",
    'A relative of European pear varieties like Bartlett and Anjou, the Asian pear is great used in recipes or simply eaten out of hand. It retains a crispness that works well in slaws and salads, and it holds its shape better than European pears when baked and cooked.',
]
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_128, NanoNFCorpus_128 and NanoNQ_128
• Evaluated with SparseInformationRetrievalEvaluator with these parameters:

  {
      "max_active_dims": 128
  }
  

Metric	NanoMSMARCO_128	NanoNFCorpus_128	NanoNQ_128
dot_accuracy@1	0.36	0.36	0.38
dot_accuracy@3	0.64	0.46	0.58
dot_accuracy@5	0.74	0.52	0.66
dot_accuracy@10	0.86	0.68	0.78
dot_precision@1	0.36	0.36	0.38
dot_precision@3	0.2133	0.32	0.2
dot_precision@5	0.148	0.28	0.136
dot_precision@10	0.086	0.246	0.082
dot_recall@1	0.36	0.0215	0.37
dot_recall@3	0.64	0.048	0.54
dot_recall@5	0.74	0.0637	0.62
dot_recall@10	0.86	0.0979	0.73
dot_ndcg@10	0.6055	0.2819	0.5492
dot_mrr@10	0.5249	0.4416	0.5042
dot_map@100	0.53	0.1105	0.4927
row_non_zero_mean_query	128.0	128.0	128.0
row_sparsity_mean_query	0.9688	0.9688	0.9688
row_non_zero_mean_corpus	128.0	128.0	128.0
row_sparsity_mean_corpus	0.9688	0.9688	0.9688

Sparse Nano BEIR

• Dataset: NanoBEIR_mean_128
• Evaluated with SparseNanoBEIREvaluator with these parameters:

  {
      "dataset_names": [
          "msmarco",
          "nfcorpus",
          "nq"
      ],
      "max_active_dims": 128
  }
  

Metric	Value
dot_accuracy@1	0.3667
dot_accuracy@3	0.56
dot_accuracy@5	0.64
dot_accuracy@10	0.7733
dot_precision@1	0.3667
dot_precision@3	0.2444
dot_precision@5	0.188
dot_precision@10	0.138
dot_recall@1	0.2505
dot_recall@3	0.4093
dot_recall@5	0.4746
dot_recall@10	0.5626
dot_ndcg@10	0.4789
dot_mrr@10	0.4902
dot_map@100	0.3777
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 Information Retrieval

• Datasets: NanoMSMARCO_256, NanoNFCorpus_256 and NanoNQ_256
• Evaluated with SparseInformationRetrievalEvaluator with these parameters:

  {
      "max_active_dims": 256
  }
  

Metric	NanoMSMARCO_256	NanoNFCorpus_256	NanoNQ_256
dot_accuracy@1	0.44	0.32	0.44
dot_accuracy@3	0.64	0.48	0.68
dot_accuracy@5	0.74	0.6	0.72
dot_accuracy@10	0.9	0.72	0.82
dot_precision@1	0.44	0.32	0.44
dot_precision@3	0.2133	0.34	0.2333
dot_precision@5	0.148	0.316	0.152
dot_precision@10	0.09	0.274	0.086
dot_recall@1	0.44	0.0394	0.42
dot_recall@3	0.64	0.073	0.63
dot_recall@5	0.74	0.0953	0.68
dot_recall@10	0.9	0.1342	0.77
dot_ndcg@10	0.6594	0.3211	0.6015
dot_mrr@10	0.5844	0.4373	0.565
dot_map@100	0.5877	0.1478	0.5493
row_non_zero_mean_query	256.0	256.0	256.0
row_sparsity_mean_query	0.9375	0.9375	0.9375
row_non_zero_mean_corpus	256.0	256.0	256.0
row_sparsity_mean_corpus	0.9375	0.9375	0.9375

Sparse Nano BEIR

• Dataset: NanoBEIR_mean_256
• Evaluated with SparseNanoBEIREvaluator with these parameters:

  {
      "dataset_names": [
          "msmarco",
          "nfcorpus",
          "nq"
      ],
      "max_active_dims": 256
  }
  

Metric	Value
dot_accuracy@1	0.4
dot_accuracy@3	0.6
dot_accuracy@5	0.6867
dot_accuracy@10	0.8133
dot_precision@1	0.4
dot_precision@3	0.2622
dot_precision@5	0.2053
dot_precision@10	0.15
dot_recall@1	0.2998
dot_recall@3	0.4477
dot_recall@5	0.5051
dot_recall@10	0.6014
dot_ndcg@10	0.5273
dot_mrr@10	0.5289
dot_map@100	0.4283
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

gooaq

• Dataset: gooaq at b089f72
• Size: 3,011,496 training samples
• Columns: question and answer
• Approximate statistics based on the first 1000 samples:

  	question	answer
  type	string	string
  details	min: 8 tokens mean: 11.87 tokens max: 23 tokens	min: 14 tokens mean: 60.09 tokens max: 201 tokens

• Samples:

  question	answer
  what is the difference between clay and mud mask?	The main difference between the two is that mud is a skin-healing agent, while clay is a cosmetic, drying agent. Clay masks are most useful for someone who has oily skin and is prone to breakouts of acne and blemishes.
  myki how much on card?	A full fare myki card costs $6 and a concession, seniors or child myki costs $3. For more information about how to use your myki, visit ptv.vic.gov.au or call 1800 800 007.
  how to find out if someone blocked your phone number on iphone?	If you get a notification like "Message Not Delivered" or you get no notification at all, that's a sign of a potential block. Next, you could try calling the person. If the call goes right to voicemail or rings once (or a half ring) then goes to voicemail, that's further evidence you may have been blocked.

• Loss: CSRLoss with these parameters:

  {
      "beta": 0.1,
      "gamma": 1.0,
      "loss": "SparseMultipleNegativesRankingLoss(scale=1.0, similarity_fct='dot_score')"
  }
  

Evaluation Dataset

gooaq

• Dataset: gooaq at b089f72
• Size: 1,000 evaluation samples
• Columns: question and answer
• Approximate statistics based on the first 1000 samples:

  	question	answer
  type	string	string
  details	min: 8 tokens mean: 11.88 tokens max: 22 tokens	min: 14 tokens mean: 61.03 tokens max: 127 tokens

• Samples:

  question	answer
  how do i program my directv remote with my tv?	['Press MENU on your remote.', 'Select Settings & Help > Settings > Remote Control > Program Remote.', 'Choose the device (TV, audio, DVD) you wish to program. ... ', 'Follow the on-screen prompts to complete programming.']
  are rodrigues fruit bats nocturnal?	Before its numbers were threatened by habitat destruction, storms, and hunting, some of those groups could number 500 or more members. Sunrise, sunset. Rodrigues fruit bats are most active at dawn, at dusk, and at night.
  why does your heart rate increase during exercise bbc bitesize?	During exercise there is an increase in physical activity and muscle cells respire more than they do when the body is at rest. The heart rate increases during exercise. The rate and depth of breathing increases - this makes sure that more oxygen is absorbed into the blood, and more carbon dioxide is removed from it.

• 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: 512
• per_device_eval_batch_size: 512
• learning_rate: 1e-05
• num_train_epochs: 1
• warmup_ratio: 0.1
• 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: 512
• per_device_eval_batch_size: 512
• 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: 1e-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: 0.1
• 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_128_dot_ndcg@10	NanoNFCorpus_128_dot_ndcg@10	NanoNQ_128_dot_ndcg@10	NanoBEIR_mean_128_dot_ndcg@10	NanoMSMARCO_256_dot_ndcg@10	NanoNFCorpus_256_dot_ndcg@10	NanoNQ_256_dot_ndcg@10	NanoBEIR_mean_256_dot_ndcg@10
-1	-1	-	-	0.6421	0.2724	0.5528	0.4891	0.6425	0.2985	0.6194	0.5201
0.0170	100	0.5414	-	-	-	-	-	-	-	-	-
0.0340	200	0.5387	-	-	-	-	-	-	-	-	-
0.0510	300	0.5183	-	-	-	-	-	-	-	-	-
0.0680	400	0.5215	-	-	-	-	-	-	-	-	-
0.0850	500	0.5011	-	-	-	-	-	-	-	-	-
0.1020	600	0.5	-	-	-	-	-	-	-	-	-
0.1190	700	0.4885	-	-	-	-	-	-	-	-	-
0.1360	800	0.4777	0.3915	0.6173	0.2782	0.5465	0.4807	0.6406	0.3038	0.6318	0.5254
0.1530	900	0.4793	-	-	-	-	-	-	-	-	-
0.1700	1000	0.472	-	-	-	-	-	-	-	-	-
0.1870	1100	0.4679	-	-	-	-	-	-	-	-	-
0.2040	1200	0.4666	-	-	-	-	-	-	-	-	-
0.2210	1300	0.4569	-	-	-	-	-	-	-	-	-
0.2380	1400	0.4642	-	-	-	-	-	-	-	-	-
0.2550	1500	0.4611	-	-	-	-	-	-	-	-	-
0.272	1600	0.4537	0.3851	0.6314	0.266	0.5664	0.4879	0.6451	0.3238	0.6363	0.5351
0.2890	1700	0.4554	-	-	-	-	-	-	-	-	-
0.3060	1800	0.4475	-	-	-	-	-	-	-	-	-
0.3230	1900	0.4512	-	-	-	-	-	-	-	-	-
0.3400	2000	0.4522	-	-	-	-	-	-	-	-	-
0.3570	2100	0.4475	-	-	-	-	-	-	-	-	-
0.3740	2200	0.4499	-	-	-	-	-	-	-	-	-
0.3910	2300	0.4467	-	-	-	-	-	-	-	-	-
0.4080	2400	0.4467	0.3940	0.6264	0.2643	0.5719	0.4875	0.6092	0.3350	0.6363	0.5268
0.4250	2500	0.4477	-	-	-	-	-	-	-	-	-
0.4420	2600	0.4466	-	-	-	-	-	-	-	-	-
0.4590	2700	0.4436	-	-	-	-	-	-	-	-	-
0.4760	2800	0.4434	-	-	-	-	-	-	-	-	-
0.4930	2900	0.4437	-	-	-	-	-	-	-	-	-
0.5100	3000	0.4381	-	-	-	-	-	-	-	-	-
0.5270	3100	0.4426	-	-	-	-	-	-	-	-	-
0.5440	3200	0.4461	0.3850	0.5866	0.2857	0.5567	0.4763	0.6232	0.3313	0.6220	0.5255
0.5610	3300	0.4453	-	-	-	-	-	-	-	-	-
0.5780	3400	0.4361	-	-	-	-	-	-	-	-	-
0.5950	3500	0.436	-	-	-	-	-	-	-	-	-
0.6120	3600	0.4444	-	-	-	-	-	-	-	-	-
0.6290	3700	0.4405	-	-	-	-	-	-	-	-	-
0.6460	3800	0.4346	-	-	-	-	-	-	-	-	-
0.6630	3900	0.4345	-	-	-	-	-	-	-	-	-
0.6800	4000	0.4399	0.3857	0.5963	0.2898	0.5537	0.4800	0.6479	0.3129	0.6058	0.5222
0.6970	4100	0.434	-	-	-	-	-	-	-	-	-
0.7140	4200	0.4353	-	-	-	-	-	-	-	-	-
0.7310	4300	0.4277	-	-	-	-	-	-	-	-	-
0.7480	4400	0.4361	-	-	-	-	-	-	-	-	-
0.7650	4500	0.445	-	-	-	-	-	-	-	-	-
0.7820	4600	0.4331	-	-	-	-	-	-	-	-	-
0.7990	4700	0.4329	-	-	-	-	-	-	-	-	-
0.8160	4800	0.4336	0.3827	0.5929	0.2894	0.5617	0.4813	0.6444	0.3241	0.6120	0.5268
0.8330	4900	0.4319	-	-	-	-	-	-	-	-	-
0.8501	5000	0.4342	-	-	-	-	-	-	-	-	-
0.8671	5100	0.439	-	-	-	-	-	-	-	-	-
0.8841	5200	0.434	-	-	-	-	-	-	-	-	-
0.9011	5300	0.4396	-	-	-	-	-	-	-	-	-
0.9181	5400	0.4355	-	-	-	-	-	-	-	-	-
0.9351	5500	0.4326	-	-	-	-	-	-	-	-	-
0.9521	5600	0.4304	0.3810	0.6055	0.2819	0.5492	0.4789	0.6594	0.3211	0.6015	0.5273
0.9691	5700	0.4316	-	-	-	-	-	-	-	-	-
0.9861	5800	0.427	-	-	-	-	-	-	-	-	-

• The bold row denotes the saved checkpoint.

Environmental Impact

Carbon emissions were measured using CodeCarbon.

• Energy Consumed: 0.906 kWh
• Carbon Emitted: 0.352 kg of CO2
• Hours Used: 2.117 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}
}