SentenceTransformer

This is a sentence-transformers model trained. 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

This model has been specifically finetuned for contradiction retrieval! It is therefore not suitable for regular similarity-based retrieval!

Model Description

Model Type: Sentence Transformer
Maximum Sequence Length: 8192 tokens
Output Dimensionality: 768 dimensions
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': 8192, 'do_lower_case': False, 'architecture': 'ModernBertModel'})
  (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("sentence_transformers_model_id")
# Run inference
sentences = [
    '(1) Posest je neposredna dejanska oblast nad stvarjo (neposredna posest).\n(2) Posest ima tudi tisti, ki izvršuje dejansko oblast nad stvarjo prek koga drugega, ki\nima neposredno posest iz kakršnegakoli pravnega naslova (posredna posest).',
    'Posameznik lahko isto stvar istočasno poseduje neposredno in posredno.',
    'Sodišče nikoli ne more odločati o poslu, ki presega redno upravljanje, brez soglasja vseh solastnikov.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.5774, 0.4022],
#         [0.5774, 1.0000, 0.2455],
#         [0.4022, 0.2455, 1.0000]])

Evaluation

Metrics

Semantic Similarity

Dataset: validation-dev
Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.0719
spearman_cosine	0.0193

Training Details

Training Dataset

Unnamed Dataset

Size: 1,174 training samples
Columns: anchor, positive, and label
Approximate statistics based on the first 1000 samples:
anchor positive label
type string string float
details
min: 2 tokens
mean: 152.19 tokens
max: 1336 tokens

min: 13 tokens
mean: 63.45 tokens
max: 304 tokens

min: 1.0
mean: 1.0
max: 1.0

	anchor	positive	label
type	string	string	float
details	min: 2 tokens mean: 152.19 tokens max: 1336 tokens	min: 13 tokens mean: 63.45 tokens max: 304 tokens	min: 1.0 mean: 1.0 max: 1.0

Samples:

anchor	positive	label
(1) Vsaka stranka v dvostranski pogodbi lahko prenese pogodbo nekomu tretjemu, ki postane s tem imetnik vseh njenih pravic in obveznosti iz te pogodbe, če v to privoli druga stranka. (2) S prenosom pogodbe preide pogodbeno razmerje med prenositeljem in drugo stranko na prevzemnika in drugo stranko takrat, ko druga stranka privoli v prenos; če je dala svojo privolitev vnaprej, pa takrat, ko je o prenosu obveščena. (3) Privolitev v prenos pogodbe je veljavna samo, če je dana v obliki, ki jo predpisuje zakon za sklenitev prenesene pogodbe. (4) Določbe o stranskih pravicah v zvezi s pogodbo o prevzemu dolga se smiselno uporabljajo tudi za prenos pogodbe.	`S prenosom pogodbe preide pogodbeno razmerje, pri čemer prenositelj jamči, da bo druga stranka izpolnila svoje obveznosti.`	`1.0`
(1) Vsaka stranka v dvostranski pogodbi lahko prenese pogodbo nekomu tretjemu, ki postane s tem imetnik vseh njenih pravic in obveznosti iz te pogodbe, če v to privoli druga stranka. (2) S prenosom pogodbe preide pogodbeno razmerje med prenositeljem in drugo stranko na prevzemnika in drugo stranko takrat, ko druga stranka privoli v prenos; če je dala svojo privolitev vnaprej, pa takrat, ko je o prenosu obveščena. (3) Privolitev v prenos pogodbe je veljavna samo, če je dana v obliki, ki jo predpisuje zakon za sklenitev prenesene pogodbe. (4) Določbe o stranskih pravicah v zvezi s pogodbo o prevzemu dolga se smiselno uporabljajo tudi za prenos pogodbe.	`S prenosom pogodbe preide pogodbeno razmerje, pri čemer prenositelj jamči, da bo druga stranka izpolnila svoje obveznosti.`	`1.0`
(1) Za škodo, ki jo povzroči delavec pri delu ali v zvezi z delom tretji osebi, odgovarja pravna ali fizična oseba, pri kateri je delavec delal takrat, ko je bila škoda povzročena, razen če dokaže, da je delavec v danih okoliščinah ravnal tako, kot je bilo treba. (2) Oškodovanec ima pravico zahtevati povrnitev škode tudi neposredno od delavca, če je ta škodo povzročil namenoma. (3) Kdor je oškodovancu povrnil škodo, ki jo je povzročil delavec namenoma ali iz hude malomarnosti, ima pravico zahtevati od delavca povrnitev plačanega zneska. (4) Ta pravica zastara v šestih mesecih od dneva, ko je bila odškodnina plačana. (5) Določba prvega odstavka tega člena ne posega v pravila o odgovornosti za škodo, ki izvira od nevarne stvari ali nevarne dejavnosti.	`Oškodovanec lahko vedno zahteva odškonino tako od delodajalca kot od delavca, njuna odgovornost je solidarna.`	`1.0`

Loss: MatryoshkaLoss with these parameters:

{
    "loss": "MultipleNegativesRankingLoss",
    "matryoshka_dims": [
        768
    ],
    "matryoshka_weights": [
        1
    ],
    "n_dims_per_step": -1
}

Evaluation Dataset

Unnamed Dataset

Size: 62 evaluation samples
Columns: anchor, positive, and label
Approximate statistics based on the first 62 samples:
anchor positive label
type string string float
details
min: 2 tokens
mean: 105.61 tokens
max: 313 tokens

min: 22 tokens
mean: 63.13 tokens
max: 187 tokens

min: 1.0
mean: 1.0
max: 1.0

	anchor	positive	label
type	string	string	float
details	min: 2 tokens mean: 105.61 tokens max: 313 tokens	min: 22 tokens mean: 63.13 tokens max: 187 tokens	min: 1.0 mean: 1.0 max: 1.0

Samples:

anchor	positive	label
`(1) Posest je neposredna dejanska oblast nad stvarjo (neposredna posest). (2) Posest ima tudi tisti, ki izvršuje dejansko oblast nad stvarjo prek koga drugega, ki ima neposredno posest iz kakršnegakoli pravnega naslova (posredna posest).`	`Posameznik lahko isto stvar istočasno poseduje neposredno in posredno.`	`1.0`
`(1) Posest je neposredna dejanska oblast nad stvarjo (neposredna posest). (2) Posest ima tudi tisti, ki izvršuje dejansko oblast nad stvarjo prek koga drugega, ki ima neposredno posest iz kakršnegakoli pravnega naslova (posredna posest).`	`Posameznik lahko isto stvar istočasno poseduje neposredno in posredno.`	`1.0`
`Upravljavec videonadzornega sistema, ki izvaja videonadzor javnih površin, mora v primeru, ko videonadzorni sistem posname dogodek, ki ogroža zdravje ali življenje posameznika, o tem nemudoma obvestiti policijo ali drug pristojni subjekt.`	`Upravljavec videonadzora ni dolžan obvestiti policije, če posnetek pokaže nevaren dogodek.`	`1.0`

Loss: MatryoshkaLoss with these parameters:

{
    "loss": "MultipleNegativesRankingLoss",
    "matryoshka_dims": [
        768
    ],
    "matryoshka_weights": [
        1
    ],
    "n_dims_per_step": -1
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 16
per_device_eval_batch_size: 16
warmup_ratio: 0.1
bf16: True
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 16
per_device_eval_batch_size: 16
gradient_accumulation_steps: 1
eval_accumulation_steps: None
torch_empty_cache_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: 3
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
use_cpu: False
seed: 42
data_seed: None
jit_mode_eval: False
bf16: True
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
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_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
parallelism_config: None
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_torch_fused
optim_args: None
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
hub_revision: None
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
include_for_metrics: []
eval_do_concat_batches: True
mp_parameters:
auto_find_batch_size: False
full_determinism: False
ray_scope: last
ddp_timeout: 1800
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
include_tokens_per_second: False
include_num_input_tokens_seen: no
neftune_noise_alpha: None
optim_target_modules: None
batch_eval_metrics: False
eval_on_start: False
use_liger_kernel: False
liger_kernel_config: None
eval_use_gather_object: False
average_tokens_across_devices: True
prompts: None
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss	validation-dev_spearman_cosine
0.3378	25	-	0.3447	0.0390
0.6757	50	-	0.2570	-0.0466
1.0135	75	-	0.2282	-0.0269
1.3514	100	0.3073	0.1797	0.0677
1.6892	125	-	0.2085	0.0184
2.0270	150	-	0.1725	0.0479
2.3649	175	-	0.1636	0.0183
2.7027	200	0.0371	0.1707	0.0193

Framework Versions

Python: 3.10.12
Sentence Transformers: 5.1.2
Transformers: 4.57.0.dev0
PyTorch: 2.9.0+cu128
Accelerate: 1.10.1
Datasets: 4.1.1
Tokenizers: 0.22.1

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",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning},
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

MultipleNegativesRankingLoss

@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}
}

Downloads last month: 6

Safetensors

Model size

0.1B params

Tensor type

F32

Papers for labpt/SLawEmbed

Evaluation results

Pearson Cosine on validation dev
self-reported

0.072
Spearman Cosine on validation dev
self-reported

0.019