This is a sentence-transformers model trained. It maps sentences & paragraphs to a 384-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

Maximum Sequence Length: 256 tokens
Output Dimensionality: 384 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': 256, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, 'pooling_mode_cls_token': False, '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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("pankajrajdeo/BioForge-bioformer-16L-umls-integration")
# Run inference
sentences = [
    'Congenital fibrinogen abnormality',
    'Congenital disease',
    'An application of magnetic resonance imaging that uses spin refocusing and spin echo generation, resulting in shorter repetition times and faster imaging.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

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

Evaluation

Metrics

Information Retrieval

Dataset: umls_sota_eval
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.9579
cosine_accuracy@3	0.9792
cosine_accuracy@5	0.9829
cosine_accuracy@10	0.9886
cosine_precision@1	0.9579
cosine_precision@3	0.5356
cosine_precision@5	0.3658
cosine_precision@10	0.206
cosine_recall@1	0.6671
cosine_recall@3	0.8833
cosine_recall@5	0.9236
cosine_recall@10	0.9553
cosine_ndcg@10	0.9525
cosine_mrr@10	0.9692
cosine_map@100	0.9383

Training Details

Training Dataset

Unnamed Dataset

Size: 2,945,832 training samples
Columns: anchor and positive
Approximate statistics based on the first 1000 samples:
anchor positive
type string string
details
min: 3 tokens
mean: 12.11 tokens
max: 62 tokens
min: 3 tokens
mean: 39.97 tokens
max: 256 tokens

Samples:

anchor	positive
`Cranial nerve structure`	`Cranial neuropathy due to petrous infection`
`Phenylalanine racemase (ATP-hydrolysing)`	`Phenylalanine racemase (adenosine triphosphate-hydrolysing) (substance)`
`Denibulin Hydrochloride`	The hydrochloride salt of denibulin, a small molecular vascular disrupting agent, with potential antimitotic and antineoplastic activities. Denibulin selectively targets and reversibly binds to the colchicine-binding site on tubulin and inhibits microtubule assembly. This results in the disruption of the cytoskeleton of tumor endothelial cells, ultimately leading to cell cycle arrest, blockage of cell division and apoptosis. This causes inadequate blood flow to the tumor and eventually leads to a decrease in tumor cell proliferation., a small molecule vascular disrupting agent (VDA), with potential antimitotic and antineoplastic activity. Denibulin selectively targets and reversibly binds to the colchicine-binding site on tubulin and inhibits microtubule assembly. This results in the disruption of the cytoskeleton of tumor endothelial cells (EC), ultimately leading to cell cycle arrest, blockage of cell division and apoptosis. This causes inadequate blood flow to the tumor and eventual...

Loss: [object Object].MultipleNegativesSymmetricMarginLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim" 4}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 512
gradient_accumulation_steps: 4
learning_rate: 1.5e-05
num_train_epochs: 4
lr_scheduler_type: cosine
warmup_ratio: 0.05
bf16: True
dataloader_num_workers: 16
load_best_model_at_end: True
gradient_checkpointing: 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: 512
per_device_eval_batch_size: 8
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 4
eval_accumulation_steps: None
torch_empty_cache_steps: None
learning_rate: 1.5e-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: 0.05
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: 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: 16
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
hub_revision: None
gradient_checkpointing: True
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
liger_kernel_config: None
eval_use_gather_object: False
average_tokens_across_devices: False
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	umls_sota_eval_cosine_ndcg@10
0.0695	100	0.8266	-
0.1390	200	0.5384	-
0.2086	300	0.4742	-
0.2781	400	0.4355	-
0.3295	474	-	0.9295
0.3476	500	0.4137	-
0.4171	600	0.3961	-
0.4866	700	0.3817	-
0.5561	800	0.3739	-
0.6257	900	0.3564	-
0.6590	948	-	0.9384
0.6952	1000	0.3587	-
0.7647	1100	0.3525	-
0.8342	1200	0.3463	-
0.9037	1300	0.3395	-
0.9732	1400	0.3329	-
0.9885	1422	-	0.9434
1.0424	1500	0.3228	-
1.1119	1600	0.318	-
1.1814	1700	0.3141	-
1.2510	1800	0.3101	-
1.3177	1896	-	0.9463
1.3205	1900	0.3134	-
1.3900	2000	0.3097	-
1.4595	2100	0.3006	-
1.5290	2200	0.303	-
1.5985	2300	0.3003	-
1.6472	2370	-	0.9484
1.6681	2400	0.2949	-
1.7376	2500	0.2951	-
1.8071	2600	0.2939	-
1.8766	2700	0.2908	-
1.9461	2800	0.2912	-
1.9767	2844	-	0.9502
2.0153	2900	0.2869	-
2.0848	3000	0.2807	-
2.1543	3100	0.2771	-
2.2238	3200	0.2795	-
2.2934	3300	0.2756	-
2.3059	3318	-	0.9510
2.3629	3400	0.2758	-
2.4324	3500	0.2765	-
2.5019	3600	0.2752	-
2.5714	3700	0.2745	-
2.6354	3792	-	0.9515
2.6409	3800	0.2714	-
2.7105	3900	0.2732	-
2.7800	4000	0.2735	-
2.8495	4100	0.2722	-
2.9190	4200	0.2713	-
2.9649	4266	-	0.9520
2.9885	4300	0.2721	-
3.0577	4400	0.2662	-
3.1272	4500	0.2654	-
3.1967	4600	0.2683	-
3.2662	4700	0.2687	-
3.2941	4740	-	0.9523
3.3358	4800	0.2665	-
3.4053	4900	0.2686	-
3.4748	5000	0.2612	-
3.5443	5100	0.263	-
3.6138	5200	0.264	-
3.6236	5214	-	0.9523
3.6834	5300	0.2672	-
3.7529	5400	0.2674	-
3.8224	5500	0.2631	-
3.8919	5600	0.2631	-
3.9531	5688	-	0.9525
3.9614	5700	0.2642	-

Framework Versions

Python: 3.11.11
Sentence Transformers: 3.4.1
Transformers: 4.53.2
PyTorch: 2.6.0+cu124
Accelerate: 1.5.2
Datasets: 3.2.0
Tokenizers: 0.21.0

Citation

BibTeX

Sentence Transformers

bibtex

@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",
}

	anchor	positive
type	string	string
details	min: 3 tokens mean: 12.11 tokens max: 62 tokens	min: 3 tokens mean: 39.97 tokens max: 256 tokens