This is a sentence-transformers model finetuned from sentence-transformers/stsb-distilbert-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: sentence-transformers/stsb-distilbert-base
Maximum Sequence Length: 128 tokens
Output Dimensionality: 768 tokens

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': 128, 'do_lower_case': False}) with Transformer model: DistilBertModel 
  (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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("tomaarsen/stsb-distilbert-base-quora-duplicate-questions")
# Run inference
sentences = [
    "What is a fetish?",
    "What's a fetish?",
    "Is it good to read sex stories?",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

Evaluation

Metrics

Binary Classification

Evaluated with BinaryClassificationEvaluator

Metric	Value
cosine_accuracy	0.7707
cosine_accuracy_threshold	0.817
cosine_f1	0.7086
cosine_f1_threshold	0.742
cosine_precision	0.6033
cosine_recall	0.8586
cosine_ap	0.7191
manhattan_accuracy	0.7729
manhattan_accuracy_threshold	181.4664
manhattan_f1	0.7083
manhattan_f1_threshold	222.9119
manhattan_precision	0.6063
manhattan_recall	0.8515
manhattan_ap	0.7188
euclidean_accuracy	0.7736
euclidean_accuracy_threshold	8.3566
euclidean_f1	0.7088
euclidean_f1_threshold	10.0929
euclidean_precision	0.6079
euclidean_recall	0.8499
euclidean_ap	0.7191
dot_accuracy	0.7442
dot_accuracy_threshold	168.5663
dot_f1	0.6832
dot_f1_threshold	142.4585
dot_precision	0.5665
dot_recall	0.8603
dot_ap	0.6694
max_accuracy	0.7736
max_accuracy_threshold	181.4664
max_f1	0.7088
max_f1_threshold	222.9119
max_precision	0.6079
max_recall	0.8603
max_ap	0.7191

Paraphrase Mining

Dataset: dev
Evaluated with ParaphraseMiningEvaluator

Metric	Value
average_precision	0.478
f1	0.5119
precision	0.4683
recall	0.5645
threshold	0.8193

Information Retrieval

Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.9654
cosine_accuracy@3	0.9904
cosine_accuracy@5	0.9948
cosine_accuracy@10	0.9974
cosine_precision@1	0.9654
cosine_precision@3	0.4355
cosine_precision@5	0.2806
cosine_precision@10	0.1493
cosine_recall@1	0.8251
cosine_recall@3	0.9549
cosine_recall@5	0.9758
cosine_recall@10	0.9898
cosine_ndcg@10	0.9786
cosine_mrr@10	0.9786
cosine_map@100	0.9714
dot_accuracy@1	0.9512
dot_accuracy@3	0.985
dot_accuracy@5	0.9914
dot_accuracy@10	0.9964
dot_precision@1	0.9512
dot_precision@3	0.4303
dot_precision@5	0.2788
dot_precision@10	0.149
dot_recall@1	0.8119
dot_recall@3	0.946
dot_recall@5	0.9708
dot_recall@10	0.9884
dot_ndcg@10	0.9703
dot_mrr@10	0.9693
dot_map@100	0.96

Training Details

Training Dataset

Unnamed Dataset

Size: 207,326 training samples
Columns: sentence_0, sentence_1, and label
Approximate statistics based on the first 1000 samples:
sentence_0 sentence_1 label
type string string int
details
min: 6 tokens
mean: 13.75 tokens
max: 42 tokens
min: 6 tokens
mean: 13.74 tokens
max: 44 tokens
1: ~100.00%

Samples:

sentence_0	sentence_1	label
`How do I improve writing skill by myself?`	`How can I improve writing skills?`	`1`
`Is it best to switch to Node.js from PHP?`	`Should I switch to Node.js or continue using PHP?`	`1`
`What do Hillary Clinton's supporters say when confronted with all her lies and scandals?`	`What do Clinton supporters say when confronted with her scandals such as the emails and 'Clinton Cash'?`	`1`

Loss: sentence_transformers.losses.MultipleNegativesRankingLoss.MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim" 4}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 64
per_device_eval_batch_size: 64
num_train_epochs: 1
round_robin_sampler: True

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
prediction_loss_only: False
per_device_train_batch_size: 64
per_device_eval_batch_size: 64
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
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}
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: None
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
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
round_robin_sampler: True

Training Logs

Epoch	Step	Training Loss	cosine_accuracy	cosine_map@100	dev_average_precision
0	0	-	0.7661	0.9371	0.4137
0.1543	500	0.1055	0.7632	0.9620	0.4731
0.3086	1000	0.0677	0.7608	0.9675	0.4732
0.4630	1500	0.0612	0.7663	0.9710	0.4856
0.6173	2000	0.0584	0.7719	0.9693	0.4925
0.7716	2500	0.0506	0.7714	0.9709	0.4808
0.9259	3000	0.0488	0.7708	0.9713	0.4784
1.0	3240	-	0.7707	0.9714	0.4780

Framework Versions

Python: 3.11.6
Sentence Transformers: 2.7.0.dev0
Transformers: 4.39.3
PyTorch: 2.1.0+cu121
Accelerate: 0.26.1
Datasets: 2.18.0
Tokenizers: 0.15.2

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",
}

MultipleNegativesRankingLoss

bibtex

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply}, 
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

	sentence_0	sentence_1	label
type	string	string	int
details	min: 6 tokens mean: 13.75 tokens max: 42 tokens	min: 6 tokens mean: 13.74 tokens max: 44 tokens	1: ~100.00%