ClementineBleuze/SBERT-roberta_pf

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("SBERT-roberta_pf")
# Run inference
sentences = [
    'section: Experimental Setup, text: The training data used for speech recognition -CSR -is different from the Treebank in two aspects: • the Treebank is only a subset of the usual CSR training data; • the Treebank tokenization is different from that of the CSR corpus; among other spurious small differences, the most frequent ones are of the type presented in',
    'section: Multiple choice next sentence prediction (NSP), text: We have collected a new dataset with 54k multiple choice questions where the objective is to predict the correct continuation for a given context sentence from four possible answer choices.',
    'section: Comparing with Previous Latent Semantic Models, text: 𝐹 𝑖 (or its translation candidate 𝐸), and 𝐲 be the projected feature vector, i.e., 𝐲 = 𝐖 T 𝐱.',
]
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]

Training Details

Training Dataset

Unnamed Dataset

• Size: 11,354 training samples
• Columns: text and label
• Approximate statistics based on the first 1000 samples:

  	text	label
  type	string	int
  details	min: 11 tokens mean: 38.95 tokens max: 234 tokens	0: ~14.30% 1: ~10.70% 2: ~23.90% 3: ~0.40% 4: ~2.60% 5: ~1.50% 6: ~1.90% 7: ~44.70%

• Samples:

  text	label
  section: INTRODUCTION, text: Arguments for the importance of prosody in language abound in the literature.	0
  section: Results, text: This overlap ensures that actions that might otherwise occur on clip boundaries will also occur as part of a clip.	7
  section: Introduction, text: In Section 4 the experimental setup and results are detailed.	6

• Loss: BatchAllTripletLoss

Evaluation Dataset

Unnamed Dataset

• Size: 1,419 evaluation samples
• Columns: text and label
• Approximate statistics based on the first 1000 samples:

  	text	label
  type	string	int
  details	min: 10 tokens mean: 40.49 tokens max: 221 tokens	0: ~15.50% 1: ~11.10% 2: ~20.50% 3: ~0.50% 4: ~2.60% 5: ~1.50% 6: ~2.10% 7: ~46.20%

• Samples:

  text	label
  section: Introduction, text: It is common in Natural Language Processing (NLP) that the categories into which text is classified do not have fully objective definitions.	0
  section: Automatic Evaluation Results, text: With respect to the BLEU score, this difference is 1.58 points absolute for the word based evaluation (27% relative increase), and 2.47 points absolute for the morphemebased evaluation (21% relative increase).	2
  section: Neural Descriptor Fields, text: The sum of the maxpooled mask probabilities of all slots can be used for counting, and the loss can be back propagated to optimize NDF as well as the embeddings.	7

• Loss: BatchAllTripletLoss

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• learning_rate: 1e-05
• weight_decay: 0.1
• load_best_model_at_end: True
• push_to_hub: 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
• learning_rate: 1e-05
• weight_decay: 0.1
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 3
• 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: 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: True
• 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: proportional

Training Logs

Epoch	Step	Training Loss	loss
0.3521	500	4.3466	4.0196
0.7042	1000	3.9809	3.3573
1.0563	1500	3.8231	3.7082
1.4085	2000	3.5722	3.6799
1.7606	2500	3.6224	3.4086
2.1127	3000	3.1266	3.2109
2.4648	3500	3.1252	3.3558
2.8169	4000	3.1115	3.1682

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.9.2
• Sentence Transformers: 3.0.1
• Transformers: 4.41.2
• PyTorch: 2.3.1+cu121
• 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",
}

BatchAllTripletLoss

@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}
}