OverSamu/reranker-sapbert-ncbi-disease-bce-context

Cambridge-SapBERT-from-PubMedBERT-fulltext trained on NCBI Disease

This is a Cross Encoder model finetuned from cambridgeltl/SapBERT-from-PubMedBERT-fulltext using the sentence-transformers library. It computes scores for pairs of texts, which can be used for text reranking and semantic search.

Model Details

Model Description

• Model Type: Cross Encoder
• Base model: cambridgeltl/SapBERT-from-PubMedBERT-fulltext
• Maximum Sequence Length: 512 tokens
• Number of Output Labels: 1 label
• Language: en

Model Sources

• Documentation: Sentence Transformers Documentation
• Documentation: Cross Encoder Documentation
• Repository: Sentence Transformers on GitHub
• Hugging Face: Cross Encoders on Hugging Face

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 CrossEncoder

# Download from the 🤗 Hub
model = CrossEncoder("cross_encoder_model_id")
# Get scores for pairs of texts
pairs = [
    ['Original mention: L-N.\nContext: Partial HPRT deficiencies are associated with gouty arthritis, while absence of activity results in Lesch-Nyhan syndrome (L-N).', 'guanine phosphoribosyltransferase deficiencies'],
    ['Original mention: breast cancer.\nContext: Most multiple case families of young onset breast cancer and ovarian cancer are thought to be due to highly penetrant mutations in the predisposing genes BRCA1 and BRCA2.', 'malignancies'],
    ['Original mention: AS.\nContext: We report here the characterization of a transgene insertion (Epstein-Barr virus Latent Membrane Protein 2A, LMP2A) into mouse chromosome 7C, which has resulted in mouse models for PWS and AS dependent on the sex of the transmitting parent.', 'achondroplastic dwarfism'],
    ['Original mention: adenomatous polyposis coli.\nContext: Epidemiologic studies have shown an increased frequency of this tumor type in families affected by adenomatous polyposis coli.', 'apc (adenomatous polyposis coli)'],
    ['Original mention: autosomal recessive disorder.\nContext: Mucopolysaccharidosis IVA (MPS IVA) is an autosomal recessive disorder caused by a deficiency in N-acetylgalactosamine-6-sulfatase (GALNS).', 'susceptibility, disease'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'Original mention: L-N.\nContext: Partial HPRT deficiencies are associated with gouty arthritis, while absence of activity results in Lesch-Nyhan syndrome (L-N).',
    [
        'guanine phosphoribosyltransferase deficiencies',
        'malignancies',
        'achondroplastic dwarfism',
        'apc (adenomatous polyposis coli)',
        'susceptibility, disease',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Reranking

• Dataset: ncbi-disease-dev
• Evaluated with CrossEncoderRerankingEvaluator with these parameters:

  {
      "at_k": 10,
      "always_rerank_positives": false
  }
  

Metric	Value
map	0.9979 (+0.5536)
mrr@10	0.9986 (+0.7236)
ndcg@10	0.9987 (+0.4194)

Training Details

Training Dataset

Unnamed Dataset

• Size: 222,348 training samples
• Columns: query, answer, and label
• Approximate statistics based on the first 1000 samples:

  	query	answer	label
  type	string	string	int
  details	min: 64 characters mean: 207.06 characters max: 558 characters	min: 5 characters mean: 27.99 characters max: 107 characters	0: ~46.10% 1: ~53.90%

• Samples:

  query	answer	label
  Original mention: L-N. Context: Partial HPRT deficiencies are associated with gouty arthritis, while absence of activity results in Lesch-Nyhan syndrome (L-N).	guanine phosphoribosyltransferase deficiencies	1
  Original mention: breast cancer. Context: Most multiple case families of young onset breast cancer and ovarian cancer are thought to be due to highly penetrant mutations in the predisposing genes BRCA1 and BRCA2.	malignancies	0
  Original mention: AS. Context: We report here the characterization of a transgene insertion (Epstein-Barr virus Latent Membrane Protein 2A, LMP2A) into mouse chromosome 7C, which has resulted in mouse models for PWS and AS dependent on the sex of the transmitting parent.	achondroplastic dwarfism	0

• Loss: BinaryCrossEntropyLoss with these parameters:

  {
      "activation_fn": "torch.nn.modules.linear.Identity",
      "pos_weight": 0.7792357206344604
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 128
• per_device_eval_batch_size: 128
• learning_rate: 2e-05
• warmup_ratio: 0.05
• seed: 12
• bf16: True
• dataloader_num_workers: 4
• load_best_model_at_end: 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: 128
• per_device_eval_batch_size: 128
• 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: 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: 3
• 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: 12
• data_seed: None
• jit_mode_eval: 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: 4
• 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}
• 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
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	ncbi-disease-dev_ndcg@10
0.0006	1	0.5977	-
0.0863	150	0.5711	-
0.1726	300	0.3596	-
0.2589	450	0.2451	-
0.3452	600	0.1933	-
0.4315	750	0.1629	-
0.5178	900	0.1448	-
0.6041	1050	0.1271	-
0.6904	1200	0.1102	-
0.7768	1350	0.0986	-
0.8631	1500	0.0927	0.9963 (+0.4169)
0.9494	1650	0.0821	-
1.0357	1800	0.073	-
1.1220	1950	0.0641	-
1.2083	2100	0.0544	-
1.2946	2250	0.055	-
1.3809	2400	0.0556	-
1.4672	2550	0.0546	-
1.5535	2700	0.0514	-
1.6398	2850	0.0463	-
1.7261	3000	0.0416	0.9984 (+0.4190)
1.8124	3150	0.043	-
1.8987	3300	0.0433	-
1.9850	3450	0.0425	-
2.0713	3600	0.0322	-
2.1577	3750	0.0272	-
2.2440	3900	0.0273	-
2.3303	4050	0.0274	-
2.4166	4200	0.0265	-
2.5029	4350	0.0285	-
2.5892	4500	0.0249	0.9987 (+0.4194)
2.6755	4650	0.0263	-
2.7618	4800	0.0252	-
2.8481	4950	0.0256	-
2.9344	5100	0.0247	-
-1	-1	-	0.9987 (+0.4194)

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.12.12
• Sentence Transformers: 5.2.0
• Transformers: 4.57.3
• PyTorch: 2.9.1+cu130
• Accelerate: 1.12.0
• Datasets: 4.4.2
• Tokenizers: 0.22.1

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",
}