This is a sentence-transformers model finetuned from distilbert/distilbert-base-uncased on the sentence-transformers/wikipedia-sections 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: distilbert/distilbert-base-uncased
Maximum Sequence Length: 512 tokens
Output Dimensionality: 768 tokens
Similarity Function: Cosine Similarity
Training Dataset:
- sentence-transformers/wikipedia-sections
Language: en

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: 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/distilbert-base-uncased-wikipedia-sections-triplet")
# Run inference
sentences = [
    'Points awarded in the final: .',
    'Points awarded in the final:[REF] .',
    'Bishop Ludden recently implemented an innovative House Program.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

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

Evaluation

Metrics

Triplet

Dataset: wikipedia-sections-dev
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.733
dot_accuracy	0.269
manhattan_accuracy	0.726
euclidean_accuracy	0.727
max_accuracy	0.733

Triplet

Dataset: wikipedia-sections-test
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.7
dot_accuracy	0.306
manhattan_accuracy	0.706
euclidean_accuracy	0.708
max_accuracy	0.708

Training Details

Training Dataset

sentence-transformers/wikipedia-sections

Dataset: sentence-transformers/wikipedia-sections at 576bb61
Size: 10,000 training samples
Columns: anchor, positive, and negative
Approximate statistics based on the first 1000 samples:
anchor positive negative
type string string string
details
min: 7 tokens
mean: 31.65 tokens
max: 72 tokens
min: 7 tokens
mean: 31.54 tokens
max: 91 tokens
min: 8 tokens
mean: 31.52 tokens
max: 150 tokens

Samples:

anchor	positive	negative
`Bailey was educated at Ipswich School (1972-79) and at the College of St Hild and St Bede University of Durham (1979-82), where he obtained a first-class degree in Economic history.`	`He won the Cricket Society's Wetherell Award in 1979 for the best public school all-rounder and played for the NCA Young Cricketers in 1980 [REF].`	`Bailey was a Fellow of Gonville and Caius College, Cambridge, between 1986 and 1996, lecturing in history and working as Admissions' Tutor.`
`The record design and production was done by Ivan Stančić Piko and the cover was chosen to be "The Red Nude" act by Amedeo Modigliani.`	`VIS Idoli was also released as a double cassette EP with Film's Live in Kulušić EP entitled Zajedno.`	`Promotional video was recorded for "Devojko mala" as the TV stations already broadcast the video for "Malena" and "Zašto su danas devojke ljute", which had its TV premiere on the 1981 New Year's Eve as part of Rokenroler show.`
`Promotional video was recorded for "Devojko mala" as the TV stations already broadcast the video for "Malena" and "Zašto su danas devojke ljute", which had its TV premiere on the 1981 New Year's Eve as part of Rokenroler show.`	`"Dok dobuje kiša (u ritmu tam-tama)" and "Malena" appeared on Vlada Divljan's 1996 live album Odbrana i zaštita.`	`The record design and production was done by Ivan Stančić Piko and the cover was chosen to be "The Red Nude" act by Amedeo Modigliani.`

Loss: TripletLoss with these parameters:
json
1{ 2 "distance_metric": "TripletDistanceMetric.EUCLIDEAN", 3 "triplet_margin": 5 4}

Evaluation Dataset

sentence-transformers/wikipedia-sections

Dataset: sentence-transformers/wikipedia-sections at 576bb61
Size: 1,000 evaluation samples
Columns: anchor, positive, and negative
Approximate statistics based on the first 1000 samples:
anchor positive negative
type string string string
details
min: 9 tokens
mean: 29.99 tokens
max: 77 tokens
min: 8 tokens
mean: 31.02 tokens
max: 88 tokens
min: 8 tokens
mean: 30.75 tokens
max: 80 tokens

Samples:

anchor	positive	negative
`Modern airforces have become dependent on airborne radars typically carried by converted airliners and transport aircraft such as the E-3 Sentry and A-50 'Mainstay'.`	`In late 2003, the missile was offered again on the export market as the 172S-1 [REF].`	`The mockup shown in 1993 had a strong resemblance to the Buk airframe, but since the Indians became involved there have been some changes.`
`In May 2005 it was reported that there were two versions, with and without a rocket booster, with ranges of 400 km and 300 km respectively [REF].`	`Guidance is by inertial navigation until the missile is close enough to the target to use active radar for terminal homing [REF].`	`The missile resurfaced as the KS-172 in 1999,[REF] as part of a new export-led strategy[REF] whereby foreign investment in a -range export model[REF] would ultimately fund a version for the Russian airforce [REF].`
`Morris was selected in the sixth round of the 2012 NFL Draft with the 173rd overall pick by the Washington Redskins [REF].`	`The day before the season opener, coach Mike Shanahan announced that Morris would be the starting running back.`	`Despite being able to afford a new car, he still drives his 1991 Mazda 626, which he nicknamed "Bentley" [REF].`

Loss: TripletLoss with these parameters:
json
1{ 2 "distance_metric": "TripletDistanceMetric.EUCLIDEAN", 3 "triplet_margin": 5 4}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 16
per_device_eval_batch_size: 16
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: False
per_device_train_batch_size: 16
per_device_eval_batch_size: 16
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: 1.0
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
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: 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
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_sampler: batch_sampler
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	loss	wikipedia-sections-dev_max_accuracy	wikipedia-sections-test_max_accuracy
0.16	100	3.8017	3.4221	0.698	-
0.32	200	3.0703	3.3261	0.717	-
0.48	300	2.9683	3.2490	0.728	-
0.64	400	2.7731	3.2340	0.733	-
0.8	500	2.9689	3.1583	0.737	-
0.96	600	2.8955	3.1480	0.733	-
1.0	625	-	-	-	0.708

Environmental Impact

Carbon emissions were measured using CodeCarbon.

Energy Consumed: 0.009 kWh
Carbon Emitted: 0.003 kg of CO2
Hours Used: 0.045 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: 3.0.0.dev0
Transformers: 4.41.0.dev0
PyTorch: 2.3.0+cu121
Accelerate: 0.26.1
Datasets: 2.18.0
Tokenizers: 0.19.1

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",
}

TripletLoss

bibtex

@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}
}

	anchor	positive	negative
type	string	string	string
details	min: 7 tokens mean: 31.65 tokens max: 72 tokens	min: 7 tokens mean: 31.54 tokens max: 91 tokens	min: 8 tokens mean: 31.52 tokens max: 150 tokens