potsu-potsu/bge-base-biomedical-matryoshka

BGE Base Biomedical MRL

This is a sentence-transformers model finetuned from BAAI/bge-base-en-v1.5 on the json dataset. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: BAAI/bge-base-en-v1.5
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • json
• Language: en
• License: apache-2.0

Model Sources

• Documentation: Sentence Transformers Documentation
• Repository: Sentence Transformers on GitHub
• Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': True}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 768, '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): Normalize()
)

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 SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("potsu-potsu/bge-base-biomedical-matryoshka")
# Run inference
sentences = [
    'What are the effects of the deletion of all three Pcdh clusters (tricluster deletion) in mice?',
    'Multicluster Pcdh diversity is required for mouse olfactory neural circuit assembly. The vertebrate clustered protocadherin (Pcdh) cell surface proteins are encoded by three closely linked gene clusters (Pcdhα, Pcdhβ, and Pcdhγ). Although deletion of individual Pcdh clusters had subtle phenotypic consequences, the loss of all three clusters (tricluster deletion) led to a severe axonal arborization defect and loss of self-avoidance.',
    'Investigators proposed that there have been three extended periods in the evolution of gene regulatory elements. Early vertebrate evolution was characterized by regulatory gains near transcription factors and developmental genes, but this trend was replaced by innovations near extracellular signaling genes, and then innovations near posttranslational protein modifiers.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

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

Evaluation

Metrics

Information Retrieval

• Dataset: dim_768
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 768
  }
  

Metric	Value
cosine_accuracy@1	0.7525
cosine_accuracy@3	0.8628
cosine_accuracy@5	0.8996
cosine_accuracy@10	0.9222
cosine_precision@1	0.7525
cosine_precision@3	0.5974
cosine_precision@5	0.5163
cosine_precision@10	0.3977
cosine_recall@1	0.2341
cosine_recall@3	0.3974
cosine_recall@5	0.4854
cosine_recall@10	0.6062
cosine_ndcg@10	0.694
cosine_mrr@10	0.8135
cosine_map@100	0.6258

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 512
  }
  

Metric	Value
cosine_accuracy@1	0.7539
cosine_accuracy@3	0.8586
cosine_accuracy@5	0.8953
cosine_accuracy@10	0.9208
cosine_precision@1	0.7539
cosine_precision@3	0.5964
cosine_precision@5	0.5143
cosine_precision@10	0.3977
cosine_recall@1	0.2334
cosine_recall@3	0.3947
cosine_recall@5	0.4796
cosine_recall@10	0.6046
cosine_ndcg@10	0.6913
cosine_mrr@10	0.8125
cosine_map@100	0.6197

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 256
  }
  

Metric	Value
cosine_accuracy@1	0.7355
cosine_accuracy@3	0.8487
cosine_accuracy@5	0.8868
cosine_accuracy@10	0.9137
cosine_precision@1	0.7355
cosine_precision@3	0.5818
cosine_precision@5	0.5018
cosine_precision@10	0.389
cosine_recall@1	0.2279
cosine_recall@3	0.379
cosine_recall@5	0.4645
cosine_recall@10	0.5878
cosine_ndcg@10	0.6743
cosine_mrr@10	0.7975
cosine_map@100	0.6003

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 128
  }
  

Metric	Value
cosine_accuracy@1	0.7058
cosine_accuracy@3	0.8133
cosine_accuracy@5	0.8501
cosine_accuracy@10	0.8953
cosine_precision@1	0.7058
cosine_precision@3	0.5535
cosine_precision@5	0.4736
cosine_precision@10	0.3661
cosine_recall@1	0.2151
cosine_recall@3	0.3572
cosine_recall@5	0.4326
cosine_recall@10	0.5469
cosine_ndcg@10	0.6358
cosine_mrr@10	0.77
cosine_map@100	0.5566

Information Retrieval

• Dataset: dim_64
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 64
  }
  

Metric	Value
cosine_accuracy@1	0.6266
cosine_accuracy@3	0.7666
cosine_accuracy@5	0.8091
cosine_accuracy@10	0.86
cosine_precision@1	0.6266
cosine_precision@3	0.5002
cosine_precision@5	0.4291
cosine_precision@10	0.3313
cosine_recall@1	0.1885
cosine_recall@3	0.3176
cosine_recall@5	0.3874
cosine_recall@10	0.4973
cosine_ndcg@10	0.5709
cosine_mrr@10	0.7041
cosine_map@100	0.4804

Training Details

Training Dataset

json

• Dataset: json
• Size: 4,012 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 5 tokens mean: 16.13 tokens max: 49 tokens	min: 3 tokens mean: 63.38 tokens max: 485 tokens

• Samples:

  anchor	positive
  What is the implication of histone lysine methylation in medulloblastoma?	Aberrant patterns of H3K4, H3K9, and H3K27 histone lysine methylation were shown to result in histone code alterations, which induce changes in gene expression, and affect the proliferation rate of cells in medulloblastoma.
  What is the role of STAG1/STAG2 proteins in differentiation?	STAG1/STAG2 proteins are tumour suppressor proteins that suppress cell proliferation and are essential for differentiation.
  What is the association between cell phone use and glioblastoma?	The association between cell phone use and incident glioblastoma remains unclear. Some studies have reported that cell phone use was associated with incident glioblastoma, and with reduced survival of patients diagnosed with glioblastoma. However, other studies have repeatedly replicated to find an association between cell phone use and glioblastoma.

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          768,
          512,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 16
• gradient_accumulation_steps: 16
• learning_rate: 2e-05
• num_train_epochs: 4
• lr_scheduler_type: cosine
• warmup_ratio: 0.1
• bf16: True
• tf32: True
• load_best_model_at_end: True
• optim: adamw_torch_fused
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: epoch
• prediction_loss_only: True
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 16
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 16
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 2e-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: 4
• max_steps: -1
• lr_scheduler_type: cosine
• 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: True
• 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_fused
• 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

Training Logs

Epoch	Step	Training Loss	dim_768_cosine_ndcg@10	dim_512_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
1.0	8	-	0.7106	0.7071	0.683	0.6384	0.5326
1.2540	10	25.4992	-	-	-	-	-
2.0	16	-	0.6976	0.6942	0.6763	0.6375	0.5635
2.5079	20	11.3871	-	-	-	-	-
3.0	24	-	0.6940	0.6907	0.6745	0.6365	0.5697
3.7619	30	8.6795	-	-	-	-	-
4.0	32	-	0.6940	0.6913	0.6743	0.6358	0.5709

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.12.5
• Sentence Transformers: 4.1.0
• Transformers: 4.52.4
• PyTorch: 2.7.1+cu128
• Accelerate: 1.7.0
• Datasets: 3.6.0
• 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",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning},
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

MultipleNegativesRankingLoss

@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}
}