AryehRotberg/ToS-Sentence-Transformers

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")
# Run inference
sentences = [
    'Pexgle will need to share your information, including personal information, in order to ensure the adequate performance of our contract with you.',
    'This service gives your personal data to third parties involved in its operation',
    'Extra data may be collected about you through promotions',
]
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: 150,468 training 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: 48.6 tokens max: 256 tokens	min: 6 tokens mean: 14.72 tokens max: 29 tokens	min: 4 tokens mean: 14.26 tokens max: 29 tokens

• Samples:

  anchor	positive	negative
  For all User Submissions, you hereby grant Guilded a license to translate, modify (for technical purposes, for example, making sure your content is viewable on a mobile device as well as a computer) and reproduce and otherwise act with respect to such User Submissions, in each case to enable us to operate the Services, as described in more detail below.	Copyright license limited for the purposes of that same service but transferable and sublicenseable	You are prohibited from sending chain letters, junk mail, spam or any unsolicited messages
  Our data is stored in the EU or USA with robust physical, digital, and procedural safeguards in place to protect your personal data, including the use of SSL encryption, redundant servers and data centers, and sophisticated perimeter security. We continuously audit for security vulnerabilities and make software patching a priority.	Information is provided about security practices	The service disables software that you are not licensed to use.
  No part of our Platform may be reproduced in any form or incorporated into any information retrieval system, electronic or mechanical, other than for your personal use. While using our Platform, you cannot redistribute your license (“Premium”, “Pro”, “Lite”) to anyone in any way that can make them use the features bound to your account. Unless otherwise specified, the developer tools and components, download areas, communication forums, and product information are for your personal and non-commercial use.	This service is only available for use individually and non-commercially.	Accessibility to this service is guaranteed at 99% or more

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Evaluation Dataset

Unnamed Dataset

• Size: 37,617 evaluation 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: 45.61 tokens max: 256 tokens	min: 6 tokens mean: 14.64 tokens max: 29 tokens	min: 6 tokens mean: 14.26 tokens max: 29 tokens

• Samples:

  anchor	positive	negative
  non-exclusive, worldwide right and license to use,	The service has non-exclusive use of your content	You are not being tracked
  We also reserve the right to suspend or end the Service at any time at our discretion and without notice. For example, we may suspend or terminate your use of the Service and remove Your Content if you’re not complying with these AUP Guidelines, or using the Service in a manner that may cause us legal liability, disrupt the Service, disrupt others’ use of the Service or, in our sole opinion, reason, cause harm.	Your account can be deleted or permanently suspended without prior notice and without a reason	The service claims to be CCPA compliant for California users
  ExpressVPN uses mobile identifiers to generate statistics related to the marketing channels and advertising partners through which users learned about and signed up for ExpressVPN mobile apps.	You are tracked via web beacons, tracking pixels, browser fingerprinting, and/or device fingerprinting	Your personal data is used for advertising

• 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}
• 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
• 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_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

Epoch	Step	Training Loss	Validation Loss	all-nli-dev_cosine_accuracy
-1	-1	-	-	0.9527
0.0106	100	1.3092	1.1396	0.9620
0.0213	200	1.0389	0.8936	0.9742
0.0319	300	0.8838	0.7500	0.9793
0.0425	400	0.7582	0.6477	0.9843
0.0532	500	0.6358	0.5727	0.9871
0.0638	600	0.6451	0.5158	0.9889
0.0744	700	0.4932	0.4715	0.9903
0.0851	800	0.4865	0.4355	0.9913
0.0957	900	0.4636	0.4035	0.9927
0.1063	1000	0.4406	0.3846	0.9930
0.1170	1100	0.3824	0.3691	0.9934
0.1276	1200	0.3967	0.3411	0.9944
0.1382	1300	0.3448	0.3264	0.9945
0.1489	1400	0.3372	0.3018	0.9955
0.1595	1500	0.3035	0.2941	0.9959
0.1701	1600	0.319	0.2864	0.9956
0.1808	1700	0.292	0.2743	0.9964
0.1914	1800	0.2647	0.2727	0.9965
0.2020	1900	0.2948	0.2517	0.9968
0.2127	2000	0.2583	0.2456	0.9971
0.2233	2100	0.2685	0.2352	0.9970
0.2339	2200	0.2879	0.2327	0.9969
0.2446	2300	0.2366	0.2271	0.9972
0.2552	2400	0.231	0.2164	0.9972
0.2658	2500	0.2639	0.2124	0.9973
0.2764	2600	0.2543	0.2078	0.9976
0.2871	2700	0.2261	0.2043	0.9972
0.2977	2800	0.2239	0.1976	0.9978
0.3083	2900	0.2271	0.1932	0.9977
0.3190	3000	0.2334	0.1845	0.9979
0.3296	3100	0.2021	0.1867	0.9981
0.3402	3200	0.2237	0.1762	0.9984
0.3509	3300	0.2109	0.1730	0.9983
0.3615	3400	0.2047	0.1663	0.9985
0.3721	3500	0.1904	0.1629	0.9984
0.3828	3600	0.1687	0.1643	0.9984
0.3934	3700	0.2071	0.1584	0.9984
0.4040	3800	0.1609	0.1543	0.9983
0.4147	3900	0.1862	0.1525	0.9984
0.4253	4000	0.1925	0.1504	0.9984
0.4359	4100	0.1714	0.1484	0.9985
0.4466	4200	0.2025	0.1472	0.9985
0.4572	4300	0.1427	0.1422	0.9986
0.4678	4400	0.1458	0.1401	0.9986
0.4785	4500	0.1796	0.1371	0.9985
0.4891	4600	0.1289	0.1317	0.9987
0.4997	4700	0.1427	0.1298	0.9988
0.5104	4800	0.1349	0.1313	0.9988
0.5210	4900	0.149	0.1293	0.9987
0.5316	5000	0.1633	0.1230	0.9988
0.5423	5100	0.1241	0.1240	0.9988
0.5529	5200	0.1532	0.1196	0.9988
0.5635	5300	0.1547	0.1173	0.9988
0.5742	5400	0.1652	0.1167	0.9990
0.5848	5500	0.1505	0.1120	0.9989
0.5954	5600	0.1309	0.1106	0.9990
0.6061	5700	0.1648	0.1089	0.9988
0.6167	5800	0.118	0.1070	0.9988
0.6273	5900	0.1207	0.1062	0.9988
0.6380	6000	0.1104	0.1046	0.9989
0.6486	6100	0.1262	0.1040	0.9989
0.6592	6200	0.1236	0.1008	0.9990
0.6699	6300	0.122	0.1005	0.9990
0.6805	6400	0.1244	0.1005	0.9991
0.6911	6500	0.1176	0.0998	0.9991
0.7018	6600	0.1215	0.0994	0.9991
0.7124	6700	0.1079	0.0983	0.9991
0.7230	6800	0.1099	0.0957	0.9991
0.7337	6900	0.1121	0.0950	0.9992
0.7443	7000	0.1137	0.0942	0.9992
0.7549	7100	0.1082	0.0929	0.9991
0.7656	7200	0.1047	0.0923	0.9991
0.7762	7300	0.1147	0.0904	0.9992
0.7868	7400	0.1336	0.0895	0.9991
0.7974	7500	0.1122	0.0889	0.9992
0.8081	7600	0.1126	0.0884	0.9993
0.8187	7700	0.116	0.0864	0.9992
0.8293	7800	0.0991	0.0857	0.9992
0.8400	7900	0.1091	0.0851	0.9992
0.8506	8000	0.1052	0.0846	0.9993
0.8612	8100	0.1105	0.0839	0.9992
0.8719	8200	0.1101	0.0836	0.9992
0.8825	8300	0.107	0.0832	0.9993
0.8931	8400	0.0867	0.0827	0.9993
0.9038	8500	0.0965	0.0823	0.9992
0.9144	8600	0.1108	0.0817	0.9993
0.9250	8700	0.1219	0.0814	0.9992
0.9357	8800	0.1169	0.0809	0.9992
0.9463	8900	0.0964	0.0805	0.9992
0.9569	9000	0.0939	0.0804	0.9992
0.9676	9100	0.0955	0.0803	0.9993
0.9782	9200	0.1076	0.0800	0.9993
0.9888	9300	0.1049	0.0798	0.9992
0.9995	9400	0.0826	0.0798	0.9993

Framework Versions

• Python: 3.9.19
• Sentence Transformers: 4.0.2
• Transformers: 4.48.1
• PyTorch: 2.4.1+cu124
• Accelerate: 1.6.0
• Datasets: 2.21.0
• Tokenizers: 0.21.0

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}
}