bobox/RoBERTa-base-unsupervised-TSDAE

SentenceTransformer based on FacebookAI/roberta-base

This is a sentence-transformers model finetuned from FacebookAI/roberta-base. 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: FacebookAI/roberta-base
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 tokens
• 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: RobertaModel 
  (1): Pooling({'word_embedding_dimension': 768, '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.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("bobox/RoBERTa-base-unsupervised-TSDAE")
# Run inference
sentences = [
    'beer wine both sugar alcohol excessive be a infections You also sweets, along with foods moldy cheese, if you prone.',
    "Since beer and wine both contain yeast and sugar (alcohol is sugar fermented by yeast), excessive drinking can definitely be a recipe for yeast infections. You should also go easy on sweets, along with foods like moldy cheese, mushrooms, and anything fermented if you're prone to yeast infections. 3.",
    'They began selling the plush animals to retailers rather than operating a store themselves. Today, Boyds is a publicly traded company that manufactures 18 million-20 million bears a year, all at a government-owned facility in China.',
]
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

• Dataset: sts-test
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.6886
spearman_cosine	0.6912
pearson_manhattan	0.6728
spearman_manhattan	0.6725
pearson_euclidean	0.6694
spearman_euclidean	0.6691
pearson_dot	0.1898
spearman_dot	0.1786
pearson_max	0.6886
spearman_max	0.6912

Training Details

Training Dataset

Unnamed Dataset

• Size: 300,000 training samples
• Columns: sentence_0 and sentence_1
• Approximate statistics based on the first 1000 samples:

  	sentence_0	sentence_1
  type	string	string
  details	min: 3 tokens mean: 19.88 tokens max: 54 tokens	min: 8 tokens mean: 46.45 tokens max: 157 tokens

• Samples:

  sentence_0	sentence_1
  us have across domestic shorthair, a cat pedigreed one between two breeds Unlike domestic shorthairs which come in of looks, Shorthair kittens the distinct	Most of us have either lived with or come across a domestic shorthair, a cat that closely resembles the pedigreed American Shorthair. The one difference between the two breeds: Unlike domestic shorthairs, which come in a variety of looks, the American Shorthair produces kittens with the same distinct appearance.
  much cost to get plugs normal with plugs, cost start $120 or if precious plugs are $150 to 200+ . 6 8 will price more required	how much does it cost to get your spark plugs changed? On a normal 4-cylinder engine with standard spark plugs, replacement cost can start around $120 up to $150+, or if precious metal spark plugs are required, $150 up to $200+. 6 cylinder and 8 Cylinder engines will increase in price, as more spark plugs are required.
  much my paycheck state income%, your income level not tax rate you is of just that a flat tax rate, those, it has the	how much taxes are taken out of my paycheck pa? Pennsylvania levies a flat state income tax rate of 3.07%. Therefore, your income level and filing status will not affect the income tax rate you pay at the state level. Pennsylvania is one of just eight states that has a flat income tax rate, and of those states, it has the lowest rate.

• Loss: DenoisingAutoEncoderLoss

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 12
• per_device_eval_batch_size: 12
• num_train_epochs: 1
• 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: 12
• per_device_eval_batch_size: 12
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_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: 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: 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: False
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• 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
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin

Training Logs

Epoch	Step	Training Loss	sts-test_spearman_cosine
0.02	500	7.1409	-
0.04	1000	6.207	-
0.05	1250	-	0.6399
0.06	1500	5.8038	-
0.08	2000	5.4963	-
0.1	2500	5.2609	0.6799
0.12	3000	5.0997	-
0.14	3500	5.0004	-
0.15	3750	-	0.7012
0.16	4000	4.8694	-
0.18	4500	4.7805	-
0.2	5000	4.6776	0.7074
0.22	5500	4.5757	-
0.24	6000	4.4598	-
0.25	6250	-	0.7185
0.26	6500	4.3865	-
0.28	7000	4.2692	-
0.3	7500	4.2224	0.7205
0.32	8000	4.1347	-
0.34	8500	4.0536	-
0.35	8750	-	0.7239
0.36	9000	4.0242	-
0.38	9500	4.0193	-
0.4	10000	3.9166	0.7153
0.42	10500	3.9004	-
0.44	11000	3.8372	-
0.45	11250	-	0.7141
0.46	11500	3.8037	-
0.48	12000	3.7788	-
0.5	12500	3.7191	0.7078
0.52	13000	3.7036	-
0.54	13500	3.6697	-
0.55	13750	-	0.7095
0.56	14000	3.6629	-
0.58	14500	3.639	-
0.6	15000	3.6048	0.7060
0.62	15500	3.6072	-
0.64	16000	3.574	-
0.65	16250	-	0.7056
0.66	16500	3.5423	-
0.68	17000	3.5379	-
0.7	17500	3.5222	0.6969
0.72	18000	3.5076	-
0.74	18500	3.5025	-
0.75	18750	-	0.6959
0.76	19000	3.4943	-
0.78	19500	3.475	-
0.8	20000	3.4874	0.6946
0.82	20500	3.4539	-
0.84	21000	3.4704	-
0.85	21250	-	0.6942
0.86	21500	3.4689	-
0.88	22000	3.4617	-
0.9	22500	3.4471	0.6917
0.92	23000	3.4541	-
0.94	23500	3.4394	-
0.95	23750	-	0.6915
0.96	24000	3.4505	-
0.98	24500	3.4533	-
1.0	25000	3.4574	0.6912

Framework Versions

• Python: 3.10.13
• Sentence Transformers: 3.0.1
• Transformers: 4.41.2
• PyTorch: 2.1.2
• Accelerate: 0.31.0
• Datasets: 2.19.2
• Tokenizers: 0.19.1

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",
}

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",
}