tomaarsen/bert-base-uncased-stsb-tsdae

SentenceTransformer based on google-bert/bert-base-uncased

This is a sentence-transformers model finetuned from google-bert/bert-base-uncased on the datasets-for-simcse 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: google-bert/bert-base-uncased
• Maximum Sequence Length: 75 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • datasets-for-simcse

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': 75, 'do_lower_case': False}) 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})
)

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("tomaarsen/bert-base-uncased-stsb-tsdae")
# Run inference
sentences = [
    'While the early models stayed close to their original form, eight subsequent generations varied substantially in size and styling.',
    'While the stayed close their, eight generations varied substantially in size and',
    '“ U ” cross of the river are a recent',
]
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

Semantic Similarity

• Datasets: sts-dev and sts-test
• Evaluated with EmbeddingSimilarityEvaluator

Metric	sts-dev	sts-test
pearson_cosine	0.6732	0.6425
spearman_cosine	0.6766	0.6322

Training Details

Training Dataset

datasets-for-simcse

• Dataset: datasets-for-simcse at e145e8b
• Size: 1,000,000 training samples
• Columns: text and noisy
• Approximate statistics based on the first 1000 samples:

  	text	noisy
  type	string	string
  details	min: 3 tokens mean: 27.96 tokens max: 75 tokens	min: 3 tokens mean: 17.68 tokens max: 75 tokens

• Samples:

  text	noisy
  White was born in Iver, England.	White was born in Iver,
  The common mangrove plants are "Rhizophora mucronata", "Sonneratia caseolaris", "Avicennia" spp., and "Aegiceras corniculatum".	plants are Rhizophora mucronata" "Sonneratia, spp.,".
  H3K9ac and H3K14ac have been shown to be part of the active promoter state.	H3K9ac been part of active promoter state.

• Loss: DenoisingAutoEncoderLoss

Evaluation Dataset

datasets-for-simcse

• Dataset: datasets-for-simcse at e145e8b
• Size: 1,000,000 evaluation samples
• Columns: text and noisy
• Approximate statistics based on the first 1000 samples:

  	text	noisy
  type	string	string
  details	min: 3 tokens mean: 28.12 tokens max: 75 tokens	min: 3 tokens mean: 17.79 tokens max: 66 tokens

• Samples:

  text	noisy
  Philippe Hervé (born 16 April 1959) is a French water polo player.	Philippe Hervé born April 1959 is French
  lies at the very edge of Scottish offshore waters, close to the maritime boundary with Norway.	the edge Scottish offshore waters close to maritime boundary with Norway
  The place is an exceptional example of the forced migration of convicts (Vinegar Hill rebels) and the development associated with punishment and reform, particularly convict labour and the associated coal mines.	The is an example of forced migration of convicts (Vinegar rebels and the development punishment and reform, particularly convict and the associated coal.

• Loss: DenoisingAutoEncoderLoss

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• learning_rate: 3e-05
• num_train_epochs: 1
• warmup_ratio: 0.1
• fp16: 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: 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: 3e-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: 1
• max_steps: -1
• lr_scheduler_type: linear
• 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: 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: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss	Validation Loss	sts-dev_spearman_cosine	sts-test_spearman_cosine
-1	-1	-	-	0.3173	-
0.0081	1000	7.5472	-	-	-
0.0162	2000	6.0196	-	-	-
0.0242	3000	5.4872	-	-	-
0.0323	4000	5.1452	-	-	-
0.0404	5000	4.8099	-	-	-
0.0485	6000	4.5211	-	-	-
0.0566	7000	4.2967	-	-	-
0.0646	8000	4.1411	-	-	-
0.0727	9000	4.031	-	-	-
0.0808	10000	3.9636	3.8297	0.7237	-
0.0889	11000	3.9046	-	-	-
0.0970	12000	3.8138	-	-	-
0.1051	13000	3.7859	-	-	-
0.1131	14000	3.7237	-	-	-
0.1212	15000	3.6881	-	-	-
0.1293	16000	3.6133	-	-	-
0.1374	17000	3.5777	-	-	-
0.1455	18000	3.5285	-	-	-
0.1535	19000	3.4974	-	-	-
0.1616	20000	3.4421	3.3523	0.6978	-
0.1697	21000	3.416	-	-	-
0.1778	22000	3.4143	-	-	-
0.1859	23000	3.3661	-	-	-
0.1939	24000	3.3408	-	-	-
0.2020	25000	3.3079	-	-	-
0.2101	26000	3.2873	-	-	-
0.2182	27000	3.2639	-	-	-
0.2263	28000	3.2323	-	-	-
0.2343	29000	3.2416	-	-	-
0.2424	30000	3.2117	3.1015	0.6895	-
0.2505	31000	3.1868	-	-	-
0.2586	32000	3.1576	-	-	-
0.2667	33000	3.1619	-	-	-
0.2747	34000	3.1445	-	-	-
0.2828	35000	3.1387	-	-	-
0.2909	36000	3.1159	-	-	-
0.2990	37000	3.09	-	-	-
0.3071	38000	3.0771	-	-	-
0.3152	39000	3.065	-	-	-
0.3232	40000	3.0589	2.9535	0.6885	-
0.3313	41000	3.0539	-	-	-
0.3394	42000	3.0211	-	-	-
0.3475	43000	3.0158	-	-	-
0.3556	44000	3.0172	-	-	-
0.3636	45000	2.9912	-	-	-
0.3717	46000	2.9776	-	-	-
0.3798	47000	2.9539	-	-	-
0.3879	48000	2.9753	-	-	-
0.3960	49000	2.9467	-	-	-
0.4040	50000	2.9429	2.8288	0.6830	-
0.4121	51000	2.9243	-	-	-
0.4202	52000	2.9273	-	-	-
0.4283	53000	2.9118	-	-	-
0.4364	54000	2.9068	-	-	-
0.4444	55000	2.8961	-	-	-
0.4525	56000	2.8621	-	-	-
0.4606	57000	2.8825	-	-	-
0.4687	58000	2.8466	-	-	-
0.4768	59000	2.868	-	-	-
0.4848	60000	2.8372	2.7335	0.6871	-
0.4929	61000	2.8322	-	-	-
0.5010	62000	2.8239	-	-	-
0.5091	63000	2.8148	-	-	-
0.5172	64000	2.8137	-	-	-
0.5253	65000	2.8043	-	-	-
0.5333	66000	2.7973	-	-	-
0.5414	67000	2.7739	-	-	-
0.5495	68000	2.7694	-	-	-
0.5576	69000	2.755	-	-	-
0.5657	70000	2.7846	2.6422	0.6773	-
0.5737	71000	2.7246	-	-	-
0.5818	72000	2.7438	-	-	-
0.5899	73000	2.7314	-	-	-
0.5980	74000	2.7213	-	-	-
0.6061	75000	2.7402	-	-	-
0.6141	76000	2.6955	-	-	-
0.6222	77000	2.7131	-	-	-
0.6303	78000	2.6951	-	-	-
0.6384	79000	2.6812	-	-	-
0.6465	80000	2.6844	2.5743	0.6827	-
0.6545	81000	2.665	-	-	-
0.6626	82000	2.6528	-	-	-
0.6707	83000	2.6819	-	-	-
0.6788	84000	2.6529	-	-	-
0.6869	85000	2.6665	-	-	-
0.6949	86000	2.6554	-	-	-
0.7030	87000	2.6299	-	-	-
0.7111	88000	2.659	-	-	-
0.7192	89000	2.632	-	-	-
0.7273	90000	2.6209	2.5051	0.6782	-
0.7354	91000	2.6023	-	-	-
0.7434	92000	2.6226	-	-	-
0.7515	93000	2.6057	-	-	-
0.7596	94000	2.601	-	-	-
0.7677	95000	2.5888	-	-	-
0.7758	96000	2.5811	-	-	-
0.7838	97000	2.565	-	-	-
0.7919	98000	2.5727	-	-	-
0.8	99000	2.5863	-	-	-
0.8081	100000	2.5534	2.4526	0.6799	-
0.8162	101000	2.5423	-	-	-
0.8242	102000	2.5655	-	-	-
0.8323	103000	2.5394	-	-	-
0.8404	104000	2.5217	-	-	-
0.8485	105000	2.5534	-	-	-
0.8566	106000	2.5264	-	-	-
0.8646	107000	2.5481	-	-	-
0.8727	108000	2.5508	-	-	-
0.8808	109000	2.5302	-	-	-
0.8889	110000	2.5223	2.4048	0.6771	-
0.8970	111000	2.5274	-	-	-
0.9051	112000	2.515	-	-	-
0.9131	113000	2.5088	-	-	-
0.9212	114000	2.5035	-	-	-
0.9293	115000	2.495	-	-	-
0.9374	116000	2.5066	-	-	-
0.9455	117000	2.4858	-	-	-
0.9535	118000	2.4803	-	-	-
0.9616	119000	2.506	-	-	-
0.9697	120000	2.4906	2.3738	0.6766	-
0.9778	121000	2.5027	-	-	-
0.9859	122000	2.4858	-	-	-
0.9939	123000	2.4928	-	-	-
-1	-1	-	-	-	0.6322

Environmental Impact

Carbon emissions were measured using CodeCarbon.

• Energy Consumed: 1.432 kWh
• Carbon Emitted: 0.557 kg of CO2
• Hours Used: 4.403 hours

Training Hardware

• On Cloud: No
• GPU Model: 1 x NVIDIA GeForce RTX 3090
• CPU Model: 13th Gen Intel(R) Core(TM) i7-13700K
• RAM Size: 31.78 GB

Framework Versions

• Python: 3.11.6
• Sentence Transformers: 3.4.0.dev0
• Transformers: 4.48.0.dev0
• PyTorch: 2.5.0+cu121
• Accelerate: 0.35.0.dev0
• Datasets: 2.20.0
• 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",
}

DenoisingAutoEncoderLoss

@inproceedings{wang-2021-TSDAE,
    title = "TSDAE: Using Transformer-based Sequential Denoising Auto-Encoderfor Unsupervised Sentence Embedding Learning",
    author = "Wang, Kexin and Reimers, Nils and Gurevych, Iryna",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2021",
    month = nov,
    year = "2021",
    address = "Punta Cana, Dominican Republic",
    publisher = "Association for Computational Linguistics",
    pages = "671--688",
    url = "https://arxiv.org/abs/2104.06979",
}