seanfarrell/icd11_ner_model_small

SentenceTransformer based on BAAI/bge-small-en-v1.5

This is a sentence-transformers model finetuned from BAAI/bge-small-en-v1.5. It maps sentences & paragraphs to a 384-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, classification, clustering, and more.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: BAAI/bge-small-en-v1.5
• Maximum Sequence Length: 30 tokens
• Output Dimensionality: 384 dimensions
• Similarity Function: Cosine Similarity
• Supported Modality: Text

Model Sources

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

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'BertModel'})
  (1): Pooling({'embedding_dimension': 384, 'pooling_mode': 'cls', '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("seanfarrell/icd11_ner_model_small")
# Run inference
queries = [
    'metabolic or transporter liver disease, unspecified',
]
documents = [
    'metabolic or transporter liver disease',
    'spontaneous rupture of synovium or tendon, unspecified',
    'staphyloma of cornea',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 384] [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[ 0.9840, -0.1296, -0.1292]])

Evaluation

Metrics

Information Retrieval

• Dataset: icd11-val
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.7478
cosine_accuracy@5	0.8718
cosine_accuracy@10	0.8966
cosine_precision@1	0.7478
cosine_precision@3	0.6254
cosine_precision@5	0.5259
cosine_precision@10	0.3787
cosine_recall@1	0.1589
cosine_recall@3	0.3773
cosine_recall@5	0.4796
cosine_recall@10	0.6031
cosine_ndcg@10	0.6649
cosine_mrr@10	0.8007
cosine_map@10	0.5745

Training Details

Training Dataset

Unnamed Dataset

• Size: 4,096,000 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 100 samples:

  	anchor	positive
  type	string	string
  modality	text	text
  details	min: 4 tokens mean: 11.07 tokens max: 30 tokens	min: 4 tokens mean: 11.07 tokens max: 28 tokens

• Samples:

  anchor	positive
  duodenal adenoma in familial polyposis	benign duodenal neoplasm
  congenital unilateral hip dislocation	developmental dysplasia of hip, unilateral
  postprocedural pulmonary arterial tree disorder	postprocedural pulmonary arterial tree complication

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": false,
      "directions": [
          "query_to_doc"
      ],
      "partition_mode": "joint",
      "hardness_mode": null,
      "hardness_strength": 0.0
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 512
• num_train_epochs: 1
• learning_rate: 2e-05
• lr_scheduler_type: cosine
• warmup_steps: 800
• weight_decay: 0.01
• bf16: True
• per_device_eval_batch_size: 256
• eval_on_start: True
• load_best_model_at_end: True
• dataloader_drop_last: True

All Hyperparameters

Click to expand

• per_device_train_batch_size: 512
• num_train_epochs: 1
• max_steps: -1
• learning_rate: 2e-05
• lr_scheduler_type: cosine
• lr_scheduler_kwargs: None
• warmup_steps: 800
• optim: adamw_torch_fused
• optim_args: None
• weight_decay: 0.01
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• optim_target_modules: None
• gradient_accumulation_steps: 1
• average_tokens_across_devices: True
• max_grad_norm: 1.0
• label_smoothing_factor: 0.0
• bf16: True
• fp16: False
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• use_liger_kernel: False
• liger_kernel_config: None
• use_cache: False
• neftune_noise_alpha: None
• torch_empty_cache_steps: None
• auto_find_batch_size: False
• log_on_each_node: True
• logging_nan_inf_filter: True
• include_num_input_tokens_seen: no
• log_level: passive
• log_level_replica: warning
• disable_tqdm: False
• project: huggingface
• trackio_space_id: None
• trackio_bucket_id: None
• trackio_static_space_id: None
• per_device_eval_batch_size: 256
• prediction_loss_only: True
• eval_on_start: True
• eval_do_concat_batches: True
• eval_use_gather_object: False
• eval_accumulation_steps: None
• include_for_metrics: []
• batch_eval_metrics: False
• save_only_model: False
• save_on_each_node: False
• enable_jit_checkpoint: False
• push_to_hub: False
• hub_private_repo: None
• hub_model_id: None
• hub_strategy: every_save
• hub_always_push: False
• hub_revision: None
• load_best_model_at_end: True
• ignore_data_skip: False
• restore_callback_states_from_checkpoint: False
• full_determinism: False
• seed: 42
• data_seed: None
• use_cpu: False
• 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
• dataloader_drop_last: True
• dataloader_num_workers: 0
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• dataloader_prefetch_factor: None
• dataloader_multiprocessing_context: None
• dataloader_in_order: True
• remove_unused_columns: True
• label_names: None
• train_sampling_strategy: random
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• ddp_static_graph: None
• ddp_backend: None
• ddp_timeout: 1800
• fsdp: None
• fsdp_config: None
• deepspeed: None
• debug: []
• skip_memory_metrics: True
• do_predict: False
• resume_from_checkpoint: None
• local_rank: -1
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}
• warmup_ratio: None

Training Logs

Click to expand

Epoch	Step	Training Loss	icd11-val_cosine_ndcg@10
0	0	-	0.4871
0.0063	50	2.5057	-
0.0125	100	2.1921	-
0.0187	150	1.6707	-
0.025	200	1.4270	-
0.0312	250	1.3379	-
0.0375	300	1.2811	-
0.0437	350	1.2222	-
0.05	400	1.1848	-
0.0563	450	1.1484	-
0.0625	500	1.1260	0.5411
0.0688	550	1.0856	-
0.075	600	1.0654	-
0.0813	650	1.0247	-
0.0875	700	1.0050	-
0.0938	750	0.9669	-
0.1	800	0.9515	-
0.1062	850	0.9404	-
0.1125	900	0.8906	-
0.1187	950	0.8692	-
0.125	1000	0.8413	0.5746
0.1313	1050	0.8439	-
0.1375	1100	0.8270	-
0.1437	1150	0.8128	-
0.15	1200	0.7762	-
0.1562	1250	0.7666	-
0.1625	1300	0.7555	-
0.1688	1350	0.7562	-
0.175	1400	0.7180	-
0.1812	1450	0.6945	-
0.1875	1500	0.7032	0.6002
0.1938	1550	0.6747	-
0.2	1600	0.6729	-
0.2062	1650	0.6604	-
0.2125	1700	0.6644	-
0.2188	1750	0.6556	-
0.225	1800	0.6440	-
0.2313	1850	0.6284	-
0.2375	1900	0.6105	-
0.2437	1950	0.5935	-
0.25	2000	0.5923	0.6120
0.2562	2050	0.5850	-
0.2625	2100	0.5787	-
0.2687	2150	0.5764	-
0.275	2200	0.5643	-
0.2812	2250	0.5655	-
0.2875	2300	0.5463	-
0.2938	2350	0.5414	-
0.3	2400	0.5368	-
0.3063	2450	0.5236	-
0.3125	2500	0.5109	0.6230
0.3187	2550	0.5308	-
0.325	2600	0.5000	-
0.3312	2650	0.5069	-
0.3375	2700	0.5120	-
0.3438	2750	0.4853	-
0.35	2800	0.4877	-
0.3563	2850	0.4786	-
0.3625	2900	0.4869	-
0.3688	2950	0.4865	-
0.375	3000	0.4561	0.6332
0.3812	3050	0.4640	-
0.3875	3100	0.4629	-
0.3937	3150	0.4575	-
0.4	3200	0.4493	-
0.4062	3250	0.4503	-
0.4125	3300	0.4301	-
0.4188	3350	0.4410	-
0.425	3400	0.4415	-
0.4313	3450	0.4316	-
0.4375	3500	0.4359	0.6441
0.4437	3550	0.4147	-
0.45	3600	0.4028	-
0.4562	3650	0.4002	-
0.4625	3700	0.4132	-
0.4688	3750	0.4021	-
0.475	3800	0.4058	-
0.4813	3850	0.4023	-
0.4875	3900	0.4038	-
0.4938	3950	0.4007	-
0.5	4000	0.3894	0.6501
0.5062	4050	0.3856	-
0.5125	4100	0.3879	-
0.5188	4150	0.3758	-
0.525	4200	0.3796	-
0.5312	4250	0.3890	-
0.5375	4300	0.3831	-
0.5437	4350	0.3806	-
0.55	4400	0.3781	-
0.5563	4450	0.3761	-
0.5625	4500	0.3723	0.6551
0.5687	4550	0.3603	-
0.575	4600	0.3595	-
0.5813	4650	0.3524	-
0.5875	4700	0.3695	-
0.5938	4750	0.3552	-
0.6	4800	0.3521	-
0.6062	4850	0.3459	-
0.6125	4900	0.3472	-
0.6188	4950	0.3511	-
0.625	5000	0.3537	0.6589
0.6312	5050	0.3490	-
0.6375	5100	0.3428	-
0.6438	5150	0.3492	-
0.65	5200	0.3471	-
0.6562	5250	0.3348	-
0.6625	5300	0.3404	-
0.6687	5350	0.3442	-
0.675	5400	0.3494	-
0.6813	5450	0.3306	-
0.6875	5500	0.3349	0.6598
0.6937	5550	0.3401	-
0.7	5600	0.3374	-
0.7063	5650	0.3306	-
0.7125	5700	0.3277	-
0.7188	5750	0.3274	-
0.725	5800	0.3318	-
0.7312	5850	0.3316	-
0.7375	5900	0.3273	-
0.7438	5950	0.3252	-
0.75	6000	0.3306	0.6636
0.7562	6050	0.3347	-
0.7625	6100	0.3186	-
0.7688	6150	0.3301	-
0.775	6200	0.3203	-
0.7812	6250	0.3234	-
0.7875	6300	0.3134	-
0.7937	6350	0.3175	-
0.8	6400	0.3312	-
0.8063	6450	0.3302	-
0.8125	6500	0.3189	0.6644
0.8187	6550	0.3191	-
0.825	6600	0.3265	-
0.8313	6650	0.3192	-
0.8375	6700	0.3111	-
0.8438	6750	0.3124	-
0.85	6800	0.3101	-
0.8562	6850	0.3188	-
0.8625	6900	0.3180	-
0.8688	6950	0.3153	-
0.875	7000	0.3159	0.6646
0.8812	7050	0.3132	-
0.8875	7100	0.3168	-
0.8938	7150	0.3227	-
0.9	7200	0.3121	-
0.9062	7250	0.3212	-
0.9125	7300	0.3195	-
0.9187	7350	0.3177	-
0.925	7400	0.3228	-
0.9313	7450	0.3172	-
0.9375	7500	0.3045	0.6652
0.9437	7550	0.3052	-
0.95	7600	0.3273	-
0.9563	7650	0.3163	-
0.9625	7700	0.3138	-
0.9688	7750	0.3280	-
0.975	7800	0.3074	-
0.9812	7850	0.3266	-
0.9875	7900	0.3214	-
0.9938	7950	0.3240	-
1.0	8000	0.3141	0.6649

• The bold row denotes the saved checkpoint.

Training Time

• Training: 8.9 minutes
• Evaluation: 45.4 seconds
• Total: 9.7 minutes

Framework Versions

• Python: 3.13.15
• Sentence Transformers: 5.7.0
• Transformers: 5.18.0
• PyTorch: 2.11.0+cu130
• Accelerate: 1.15.0
• Datasets: 4.8.5
• Tokenizers: 0.23.2

Additional Resources

• Training and Finetuning Embedding Models with Sentence Transformers: the end-to-end guide for training or finetuning Sentence Transformer models.
• Introduction to Matryoshka Embedding Models: variable-size embeddings that can be truncated with minimal quality loss.
• Binary and Scalar Embedding Quantization for Significantly Faster & Cheaper Retrieval: post-training compression of embedding vectors.
• Multimodal Embedding & Reranker Models with Sentence Transformers: use text, image, audio, and video models through the same API.
• Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers: train multimodal embedding models, with a Visual Document Retrieval walkthrough.

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",
}

MultipleNegativesRankingLoss

@misc{oord2019representationlearningcontrastivepredictive,
      title={Representation Learning with Contrastive Predictive Coding},
      author={Aaron van den Oord and Yazhe Li and Oriol Vinyals},
      year={2019},
      eprint={1807.03748},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1807.03748},
}