This is a sentence-transformers model finetuned from sentence-transformers/all-MiniLM-L6-v2. It maps sentences & paragraphs to a 384-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/all-MiniLM-L6-v2
Maximum Sequence Length: 256 tokens
Output Dimensionality: 384 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': 256, 'do_lower_case': False, 'architecture': 'BertModel'})
  (1): Pooling({'word_embedding_dimension': 384, '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})
  (2): Normalize()
)

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("AryehRotberg/ToS-Sentence-Transformers-V4")
# Run inference
sentences = [
    'The Services may contain links or connections to third party websites or services that are not owned or controlled by Guilded. When you access third party websites or use third party services, you accept that there are risks in doing so, and that Guilded is not responsible for such risks.',
    'This service assumes no responsibility and liability for the contents of links to other websites',
    'Copyright license limited for the purposes of that same service but transferable and sublicenseable',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[ 1.0000,  0.6397, -0.0500],
#         [ 0.6397,  1.0000,  0.0874],
#         [-0.0500,  0.0874,  1.0000]])

Evaluation

Metrics

Triplet

Dataset: all-nli-dev
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.9993

Training Details

Training Dataset

Unnamed Dataset

Size: 122,856 training samples
Columns: anchor, positive, and negative
Approximate statistics based on the first 1000 samples:
anchor positive negative
type string string string
details
min: 3 tokens
mean: 48.49 tokens
max: 256 tokens
min: 6 tokens
mean: 15.21 tokens
max: 29 tokens
min: 6 tokens
mean: 14.34 tokens
max: 29 tokens

Samples:

anchor	positive	negative
`If you ever decide to stop using Snapchat, you can just ask us to delete your account.`	`You have the right to leave this service at any time`	`Your personal information is used for many different purposes`
`you forever waive and agree not to claim or assert any entitlement to any and all moral rights of an author in any of the User Content.`	`You waive your moral rights`	`You aren’t allowed to remove or edit user-generated content`
You agree and shall indemnify and hold Dailymotion- harmless from and against any liability, loss, damages (including punitive damages), claim, settlement payment, cost and expense, interest, award, judgment, diminution in value, fine, fee (including reasonable attorneys’ fees), and penalty, or other charge (including reasonable attorneys’ fees and all other cost of investigating, defending or asserting any claim for indemnification under these Terms) arising from or relating to (i) Your Content, (ii) Your violation of the Terms or any other policy of Dailymotion. (iii) Your use of the Dailymotion Service. and (iv) Your violation of any third party rights, including without limitation any copyright, property, publicity or privacy rights.	`You agree to defend, indemnify, and hold the service harmless in case of a claim related to your use of the service`	`User-generated content can be blocked or censored for any reason`

Loss: MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim", 4 "gather_across_devices": false 5}

Evaluation Dataset

Unnamed Dataset

Size: 30,714 evaluation samples
Columns: anchor, positive, and negative
Approximate statistics based on the first 1000 samples:
anchor positive negative
type string string string
details
min: 4 tokens
mean: 49.34 tokens
max: 256 tokens
min: 6 tokens
mean: 15.13 tokens
max: 29 tokens
min: 6 tokens
mean: 14.28 tokens
max: 29 tokens

Samples:

anchor	positive	negative
`YOU AGREE THAT USE OF THE WEB SITE AND THE SERVICES IS AT YOUR SOLE RISK.`	`The service is provided 'as is' and to be used at your sole risk`	`The court of law governing the terms is in a jurisdiction that is friendlier to user privacy protection.`
`If you continue to use our services after the changes have taken effect, it means that you agree to the changes.`	`Terms may be changed at any time`	`The service is only available in some countries approved by its government`
`We may revise these Terms of Use or any of the other Terms from time to time. You are ,expected to check this page and our Terms from time to time to take notice of any changes`	`Terms may be changed at any time`	`Voice data is collected and shared with third-parties`

Loss: MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim", 4 "gather_across_devices": false 5}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 16
per_device_eval_batch_size: 16
learning_rate: 2e-05
num_train_epochs: 1
warmup_ratio: 0.1
fp16: 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
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 1
eval_accumulation_steps: None
torch_empty_cache_steps: None
learning_rate: 2e-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
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
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}
parallelism_config: None
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_torch_fused
optim_args: None
adafactor: False
group_by_length: False
length_column_name: length
project: huggingface
trackio_space_id: trackio
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_inputs_for_metrics: False
include_for_metrics: []
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
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	all-nli-dev_cosine_accuracy
-1	-1	-	-	0.9426
0.0130	100	1.4227	1.1709	0.9595
0.0260	200	1.1178	0.9104	0.9727
0.0391	300	0.9473	0.7546	0.9799
0.0521	400	0.7559	0.6471	0.9853
0.0651	500	0.6617	0.5684	0.9880
0.0781	600	0.5857	0.5047	0.9899
0.0912	700	0.5768	0.4578	0.9910
0.1042	800	0.493	0.4281	0.9921
0.1172	900	0.4877	0.3899	0.9931
0.1302	1000	0.4315	0.3593	0.9939
0.1432	1100	0.3894	0.3458	0.9940
0.1563	1200	0.3681	0.3215	0.9945
0.1693	1300	0.3533	0.3151	0.9951
0.1823	1400	0.3242	0.3093	0.9949
0.1953	1500	0.346	0.2820	0.9955
0.2084	1600	0.3212	0.2637	0.9960
0.2214	1700	0.2889	0.2601	0.9960
0.2344	1800	0.2855	0.2423	0.9960
0.2474	1900	0.2621	0.2396	0.9964
0.2605	2000	0.265	0.2299	0.9968
0.2735	2100	0.2401	0.2191	0.9969
0.2865	2200	0.254	0.2166	0.9966
0.2995	2300	0.2543	0.2036	0.9971
0.3125	2400	0.2667	0.1958	0.9973
0.3256	2500	0.2236	0.1937	0.9972
0.3386	2600	0.232	0.1875	0.9974
0.3516	2700	0.2021	0.1806	0.9977
0.3646	2800	0.2147	0.1787	0.9974
0.3777	2900	0.1929	0.1727	0.9975
0.3907	3000	0.1778	0.1721	0.9977
0.4037	3100	0.2031	0.1678	0.9974
0.4167	3200	0.1784	0.1645	0.9978
0.4297	3300	0.183	0.1593	0.9977
0.4428	3400	0.1878	0.1508	0.9979
0.4558	3500	0.1915	0.1478	0.9980
0.4688	3600	0.1611	0.1448	0.9983
0.4818	3700	0.1606	0.1385	0.9983
0.4949	3800	0.1604	0.1408	0.9984
0.5079	3900	0.1733	0.1327	0.9983
0.5209	4000	0.159	0.1277	0.9986
0.5339	4100	0.1554	0.1255	0.9987
0.5469	4200	0.1546	0.1225	0.9985
0.5600	4300	0.1536	0.1222	0.9984
0.5730	4400	0.1253	0.1174	0.9987
0.5860	4500	0.151	0.1137	0.9986
0.5990	4600	0.1293	0.1116	0.9988
0.6121	4700	0.1272	0.1093	0.9986
0.6251	4800	0.1326	0.1074	0.9985
0.6381	4900	0.135	0.1044	0.9987
0.6511	5000	0.1253	0.1013	0.9989
0.6641	5100	0.1466	0.0995	0.9989
0.6772	5200	0.1378	0.0993	0.9991
0.6902	5300	0.1245	0.0959	0.9989
0.7032	5400	0.1124	0.0946	0.9989
0.7162	5500	0.0937	0.0926	0.9988
0.7293	5600	0.1378	0.0907	0.9990
0.7423	5700	0.1234	0.0889	0.9991
0.7553	5800	0.1153	0.0876	0.9991
0.7683	5900	0.1172	0.0865	0.9990
0.7814	6000	0.1135	0.0855	0.9992
0.7944	6100	0.1178	0.0834	0.9991
0.8074	6200	0.1195	0.0812	0.9991
0.8204	6300	0.1068	0.0795	0.9991
0.8334	6400	0.0824	0.0791	0.9992
0.8465	6500	0.1173	0.0768	0.9992
0.8595	6600	0.1166	0.0757	0.9992
0.8725	6700	0.1119	0.0755	0.9992
0.8855	6800	0.1017	0.0750	0.9993
0.8986	6900	0.1148	0.0745	0.9993
0.9116	7000	0.0976	0.0736	0.9993
0.9246	7100	0.0973	0.0728	0.9993
0.9376	7200	0.0984	0.0726	0.9993
0.9506	7300	0.0943	0.0723	0.9993
0.9637	7400	0.0825	0.0719	0.9993
0.9767	7500	0.0961	0.0716	0.9993
0.9897	7600	0.0893	0.0715	0.9993

Framework Versions

Python: 3.12.11
Sentence Transformers: 5.1.1
Transformers: 4.57.0
PyTorch: 2.8.0+cu126
Accelerate: 1.10.1
Datasets: 4.0.0
Tokenizers: 0.22.1

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",
}

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}
}

	anchor	positive	negative
type	string	string	string
details	min: 3 tokens mean: 48.49 tokens max: 256 tokens	min: 6 tokens mean: 15.21 tokens max: 29 tokens	min: 6 tokens mean: 14.34 tokens max: 29 tokens