yyzheng00/sapbert_cls_lora_triplet

SentenceTransformer based on cambridgeltl/SapBERT-from-PubMedBERT-fulltext

This is a sentence-transformers model finetuned from cambridgeltl/SapBERT-from-PubMedBERT-fulltext. It maps sentences & paragraphs to a 1536-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: cambridgeltl/SapBERT-from-PubMedBERT-fulltext
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 1536 dimensions
• Similarity Function: Cosine Similarity

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': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
)

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("yyzheng00/sapbert_cls_lora_triplet")
# Run inference
sentences = [
    '|Product containing rifampicin (medicinal product)| + |Product manufactured as parenteral dose form (product)| : |Has manufactured dose form (attribute)| = |Parenteral dose form (dose form)|, { |Has active ingredient (attribute)| = |Rifampicin (substance)| }',
    'Rifampin in parenteral dosage form (medicinal product form)',
    'Product containing tocilizumab in parenteral dose form (medicinal product form)',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1536]

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

Evaluation

Metrics

Triplet

• Dataset: eval
• Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.9728

Training Details

Training Dataset

Unnamed Dataset

• Size: 600,000 training samples
• Columns: sentence_0, sentence_1, and sentence_2
• Approximate statistics based on the first 1000 samples:

  	sentence_0	sentence_1	sentence_2
  type	string	string	string
  details	min: 7 tokens mean: 46.82 tokens max: 416 tokens	min: 6 tokens mean: 12.36 tokens max: 34 tokens	min: 7 tokens mean: 21.46 tokens max: 321 tokens

• Samples:

  sentence_0	sentence_1	sentence_2
  Vassar-Culling stain method (procedure)	Vassar-Culling stain (procedure)	Durazol red stain method (procedure)
  	Product containing sodium iodide (medicinal product)	+
  	Product containing lorazepam (medicinal product)	+

• Loss: TripletLoss with these parameters:

  {
      "distance_metric": "TripletDistanceMetric.EUCLIDEAN",
      "triplet_margin": 0.2
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• num_train_epochs: 1
• fp16: True
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 8
• per_device_eval_batch_size: 8
• 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: 5e-05
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1
• 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: False
• fp16: True
• 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: False
• 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: batch_sampler
• multi_dataset_batch_sampler: round_robin

Training Logs

Click to expand

Epoch	Step	Training Loss	eval_cosine_accuracy
0.0067	500	0.2021	-
0.0133	1000	0.1845	-
0.02	1500	0.1834	-
0.0267	2000	0.1352	0.9591
0.0333	2500	0.1107	-
0.04	3000	0.1072	-
0.0467	3500	0.15	-
0.0533	4000	0.1009	0.9670
0.06	4500	0.1128	-
0.0667	5000	0.1066	-
0.0733	5500	0.0963	-
0.08	6000	0.0882	0.9682
0.0867	6500	0.0786	-
0.0933	7000	0.0803	-
0.1	7500	0.0708	-
0.1067	8000	0.0544	0.9682
0.1133	8500	0.0587	-
0.12	9000	0.0489	-
0.1267	9500	0.0418	-
0.1333	10000	0.0339	0.9669
0.14	10500	0.0358	-
0.1467	11000	0.0316	-
0.1533	11500	0.0305	-
0.16	12000	0.0233	0.9686
0.1667	12500	0.0308	-
0.1733	13000	0.0261	-
0.18	13500	0.0264	-
0.1867	14000	0.0311	0.9678
0.1933	14500	0.0241	-
0.2	15000	0.0261	-
0.2067	15500	0.0243	-
0.2133	16000	0.0252	0.9699
0.22	16500	0.0235	-
0.2267	17000	0.0226	-
0.2333	17500	0.0223	-
0.24	18000	0.0259	0.9706
0.2467	18500	0.022	-
0.2533	19000	0.0237	-
0.26	19500	0.0236	-
0.2667	20000	0.0241	0.9701
0.2733	20500	0.0235	-
0.28	21000	0.023	-
0.2867	21500	0.0235	-
0.2933	22000	0.0226	0.9717
0.3	22500	0.0225	-
0.3067	23000	0.0206	-
0.3133	23500	0.0206	-
0.32	24000	0.0208	0.9719
0.3267	24500	0.0192	-
0.3333	25000	0.0223	-
0.34	25500	0.0199	-
0.3467	26000	0.0201	0.9715
0.3533	26500	0.0166	-
0.36	27000	0.018	-
0.3667	27500	0.0201	-
0.3733	28000	0.0193	0.9722
0.38	28500	0.0228	-
0.3867	29000	0.0237	-
0.3933	29500	0.0218	-
0.4	30000	0.0205	0.9718
0.4067	30500	0.0202	-
0.4133	31000	0.0226	-
0.42	31500	0.0216	-
0.4267	32000	0.0181	0.9719
0.4333	32500	0.0214	-
0.44	33000	0.0175	-
0.4467	33500	0.0195	-
0.4533	34000	0.0189	0.9720
0.46	34500	0.0167	-
0.4667	35000	0.0175	-
0.4733	35500	0.0185	-
0.48	36000	0.0166	0.9723
0.4867	36500	0.0215	-
0.4933	37000	0.0168	-
0.5	37500	0.0154	-
0.5067	38000	0.0197	0.9700
0.5133	38500	0.0203	-
0.52	39000	0.02	-
0.5267	39500	0.0193	-
0.5333	40000	0.0156	0.9723
0.54	40500	0.0175	-
0.5467	41000	0.0168	-
0.5533	41500	0.0159	-
0.56	42000	0.02	0.9723
0.5667	42500	0.0147	-
0.5733	43000	0.0157	-
0.58	43500	0.0204	-
0.5867	44000	0.0193	0.9715
0.5933	44500	0.0167	-
0.6	45000	0.0147	-
0.6067	45500	0.0166	-
0.6133	46000	0.016	0.9721
0.62	46500	0.0166	-
0.6267	47000	0.0178	-
0.6333	47500	0.0151	-
0.64	48000	0.0172	0.9713
0.6467	48500	0.0147	-
0.6533	49000	0.0173	-
0.66	49500	0.0161	-
0.6667	50000	0.0189	0.9715
0.6733	50500	0.0186	-
0.68	51000	0.0166	-
0.6867	51500	0.0164	-
0.6933	52000	0.0188	0.9724
0.7	52500	0.0174	-
0.7067	53000	0.0166	-
0.7133	53500	0.0176	-
0.72	54000	0.0165	0.9727
0.7267	54500	0.0169	-
0.7333	55000	0.0172	-
0.74	55500	0.0166	-
0.7467	56000	0.0166	0.9729
0.7533	56500	0.016	-
0.76	57000	0.0182	-
0.7667	57500	0.0172	-
0.7733	58000	0.0173	0.9730
0.78	58500	0.0149	-
0.7867	59000	0.0159	-
0.7933	59500	0.0147	-
0.8	60000	0.0153	0.9725
0.8067	60500	0.0149	-
0.8133	61000	0.0162	-
0.82	61500	0.0154	-
0.8267	62000	0.0174	0.9725
0.8333	62500	0.0157	-
0.84	63000	0.0167	-
0.8467	63500	0.0172	-
0.8533	64000	0.0155	0.9720
0.86	64500	0.0171	-
0.8667	65000	0.0144	-
0.8733	65500	0.0144	-
0.88	66000	0.0189	0.9722
0.8867	66500	0.018	-
0.8933	67000	0.015	-
0.9	67500	0.0167	-
0.9067	68000	0.0145	0.9727
0.9133	68500	0.0165	-
0.92	69000	0.017	-
0.9267	69500	0.0145	-
0.9333	70000	0.0168	0.9726
0.94	70500	0.0133	-
0.9467	71000	0.0132	-
0.9533	71500	0.0157	-
0.96	72000	0.016	0.9729
0.9667	72500	0.0182	-
0.9733	73000	0.0159	-
0.98	73500	0.0162	-
0.9867	74000	0.0161	0.9730
0.9933	74500	0.0173	-
1.0	75000	0.0189	0.9728

Framework Versions

• Python: 3.11.1
• Sentence Transformers: 4.1.0
• Transformers: 4.47.0
• PyTorch: 2.1.1+cu121
• Accelerate: 1.2.0
• Datasets: 2.18.0
• Tokenizers: 0.21.0

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",
}

TripletLoss

@misc{hermans2017defense,
    title={In Defense of the Triplet Loss for Person Re-Identification},
    author={Alexander Hermans and Lucas Beyer and Bastian Leibe},
    year={2017},
    eprint={1703.07737},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}