AryehRotberg/ToS-Sentence-Transformers-V2

SentenceTransformer based on sentence-transformers/all-MiniLM-L6-v2

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}) with Transformer model: 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.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("AryehRotberg/ToS-Sentence-Transformers-V2")
# Run inference
sentences = [
    'Each customer may register only one Coinbase account.',
    'Alternative accounts are not allowed',
    'Usernames can be rejected or changed for any reason',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

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

Evaluation

Metrics

Triplet

• Dataset: all-nli-dev
• Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.9993

Training Details

Training Dataset

Unnamed Dataset

• Size: 203,040 training samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 5 tokens mean: 47.01 tokens max: 256 tokens	min: 6 tokens mean: 15.08 tokens max: 29 tokens	min: 4 tokens mean: 14.45 tokens max: 29 tokens

• Samples:

  anchor	positive	negative
  but remains subject to the promises made in any pre-existing Privacy Policy (unless, of course, the customer consents otherwise).	Promises will be kept after a merger or acquisition	When the service wants to change its terms, you are notified a week or more in advance.
  Visits are logged by the Web server. These logs are only used for maintenance purposes and to generate anonymous access statistics.	Only necessary logs are kept by the service to ensure quality	An onion site accessible over Tor is provided
  You affirm that you are over the age of 13, as the FanFiction.Net Service is not intended for children under 13.	This service is only available to users over a certain age	No need to register

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim"
  }
  

Evaluation Dataset

Unnamed Dataset

• Size: 50,760 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: 45.97 tokens max: 256 tokens	min: 4 tokens mean: 14.82 tokens max: 29 tokens	min: 4 tokens mean: 14.36 tokens max: 29 tokens

• Samples:

  anchor	positive	negative
  HP is not required to host, display, or distribute any User Submissions on or through This Website and may remove at any time or refuse any User Submissions for any reason.	User-generated content can be blocked or censored for any reason	The service will only respond to government requests that are reasonable
  How we use information we collect	Information is provided about how your personal data is used	The service does not index or open files that you upload
  your use of the LYKA Service is solely for your own personal use and you therefore must not, nor attempt to, resell or charge others for use of or access to the LYKA Service or for any business purposes;	This service is only available for use individually and non-commercially.	You cannot opt out of promotional communications

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim"
  }
  

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
• 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}
• tp_size: 0
• 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: False
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: None
• hub_always_push: False
• 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: False
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• eval_use_gather_object: False
• average_tokens_across_devices: False
• prompts: None
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss	Validation Loss	all-nli-dev_cosine_accuracy
-1	-1	-	-	0.9547
0.0079	100	1.3098	1.1250	0.9618
0.0158	200	1.0671	0.9039	0.9726
0.0236	300	0.8861	0.7616	0.9788
0.0315	400	0.7625	0.6672	0.9824
0.0394	500	0.7217	0.5984	0.9852
0.0473	600	0.6612	0.5432	0.9875
0.0552	700	0.5484	0.5048	0.9884
0.0630	800	0.5435	0.4699	0.9898
0.0709	900	0.522	0.4319	0.9909
0.0788	1000	0.4715	0.4152	0.9915
0.0867	1100	0.4495	0.3909	0.9923
0.0946	1200	0.4552	0.3741	0.9929
0.1024	1300	0.4159	0.3559	0.9934
0.1103	1400	0.4095	0.3404	0.9937
0.1182	1500	0.3849	0.3267	0.9936
0.1261	1600	0.3357	0.3208	0.9941
0.1340	1700	0.4029	0.2989	0.9946
0.1418	1800	0.3413	0.2882	0.9949
0.1497	1900	0.3254	0.2842	0.9952
0.1576	2000	0.3123	0.2817	0.9950
0.1655	2100	0.3003	0.2652	0.9955
0.1734	2200	0.3117	0.2559	0.9959
0.1812	2300	0.332	0.2504	0.9959
0.1891	2400	0.2923	0.2481	0.9962
0.1970	2500	0.2747	0.2389	0.9961
0.2049	2600	0.2507	0.2355	0.9962
0.2128	2700	0.2563	0.2294	0.9965
0.2206	2800	0.2512	0.2228	0.9967
0.2285	2900	0.2622	0.2201	0.9967
0.2364	3000	0.234	0.2183	0.9968
0.2443	3100	0.2607	0.2158	0.9969
0.2522	3200	0.2221	0.2077	0.9973
0.2600	3300	0.2559	0.2037	0.9971
0.2679	3400	0.2261	0.2044	0.9969
0.2758	3500	0.2453	0.1985	0.9969
0.2837	3600	0.2251	0.1927	0.9975
0.2916	3700	0.2716	0.1913	0.9976
0.2994	3800	0.1949	0.1894	0.9975
0.3073	3900	0.2361	0.1868	0.9973
0.3152	4000	0.223	0.1812	0.9974
0.3231	4100	0.1846	0.1788	0.9974
0.3310	4200	0.2143	0.1771	0.9974
0.3388	4300	0.2063	0.1705	0.9976
0.3467	4400	0.2207	0.1693	0.9977
0.3546	4500	0.2053	0.1608	0.9980
0.3625	4600	0.1705	0.1603	0.9981
0.3704	4700	0.2085	0.1597	0.9980
0.3783	4800	0.2034	0.1561	0.9981
0.3861	4900	0.1765	0.1562	0.9981
0.3940	5000	0.1955	0.1497	0.9982
0.4019	5100	0.1843	0.1487	0.9981
0.4098	5200	0.186	0.1479	0.9981
0.4177	5300	0.1631	0.1498	0.9980
0.4255	5400	0.1719	0.1468	0.9980
0.4334	5500	0.1916	0.1436	0.9983
0.4413	5600	0.1706	0.1421	0.9982
0.4492	5700	0.1512	0.1372	0.9984
0.4571	5800	0.1626	0.1357	0.9984
0.4649	5900	0.1652	0.1332	0.9985
0.4728	6000	0.146	0.1325	0.9986
0.4807	6100	0.1487	0.1308	0.9986
0.4886	6200	0.1565	0.1290	0.9985
0.4965	6300	0.1567	0.1281	0.9985
0.5043	6400	0.1678	0.1264	0.9985
0.5122	6500	0.1203	0.1261	0.9986
0.5201	6600	0.1572	0.1245	0.9985
0.5280	6700	0.1539	0.1221	0.9985
0.5359	6800	0.1546	0.1226	0.9986
0.5437	6900	0.1216	0.1185	0.9987
0.5516	7000	0.1272	0.1193	0.9986
0.5595	7100	0.1321	0.1179	0.9988
0.5674	7200	0.1305	0.1144	0.9988
0.5753	7300	0.1558	0.1151	0.9987
0.5831	7400	0.1282	0.1133	0.9986
0.5910	7500	0.1442	0.1113	0.9986
0.5989	7600	0.1529	0.1094	0.9988
0.6068	7700	0.1254	0.1086	0.9987
0.6147	7800	0.1158	0.1061	0.9988
0.6225	7900	0.1127	0.1063	0.9988
0.6304	8000	0.1253	0.1052	0.9988
0.6383	8100	0.1542	0.1050	0.9989
0.6462	8200	0.1237	0.1038	0.9990
0.6541	8300	0.1307	0.1029	0.9988
0.6619	8400	0.1231	0.1022	0.9989
0.6698	8500	0.1573	0.1002	0.9990
0.6777	8600	0.1257	0.0990	0.9990
0.6856	8700	0.103	0.0986	0.9990
0.6935	8800	0.1143	0.0983	0.9990
0.7013	8900	0.1138	0.0965	0.9991
0.7092	9000	0.1158	0.0962	0.9990
0.7171	9100	0.1104	0.0960	0.9991
0.7250	9200	0.1054	0.0967	0.9991
0.7329	9300	0.1194	0.0946	0.9991
0.7407	9400	0.1245	0.0936	0.9991
0.7486	9500	0.126	0.0926	0.9991
0.7565	9600	0.1059	0.0913	0.9992
0.7644	9700	0.1101	0.0906	0.9992
0.7723	9800	0.1192	0.0898	0.9993
0.7801	9900	0.1241	0.0886	0.9993
0.7880	10000	0.1134	0.0876	0.9993
0.7959	10100	0.1071	0.0868	0.9993
0.8038	10200	0.1043	0.0869	0.9993
0.8117	10300	0.1191	0.0864	0.9993
0.8195	10400	0.1188	0.0853	0.9993
0.8274	10500	0.1014	0.0847	0.9993
0.8353	10600	0.0878	0.0846	0.9993
0.8432	10700	0.0952	0.0839	0.9993
0.8511	10800	0.1169	0.0841	0.9993
0.8589	10900	0.1032	0.0825	0.9993
0.8668	11000	0.1086	0.0823	0.9993
0.8747	11100	0.1058	0.0820	0.9993
0.8826	11200	0.0973	0.0818	0.9993
0.8905	11300	0.1166	0.0811	0.9993
0.8983	11400	0.0965	0.0807	0.9993
0.9062	11500	0.0974	0.0805	0.9993
0.9141	11600	0.0984	0.0803	0.9993
0.9220	11700	0.1199	0.0798	0.9993
0.9299	11800	0.0854	0.0794	0.9993
0.9377	11900	0.1004	0.0798	0.9993
0.9456	12000	0.1119	0.0792	0.9993
0.9535	12100	0.1171	0.0790	0.9993
0.9614	12200	0.1045	0.0787	0.9993
0.9693	12300	0.1116	0.0784	0.9993
0.9771	12400	0.091	0.0781	0.9993
0.9850	12500	0.083	0.0781	0.9993
0.9929	12600	0.1146	0.0779	0.9993

Framework Versions

• Python: 3.11.12
• Sentence Transformers: 3.4.1
• Transformers: 4.51.3
• PyTorch: 2.6.0+cu124
• Accelerate: 1.5.2
• Datasets: 3.5.0
• Tokenizers: 0.21.1

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",
}

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}
}