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

This is a sentence-transformers model finetuned from sentence-transformers/paraphrase-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/paraphrase-MiniLM-L6-v2
Maximum Sequence Length: 128 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': 128, '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})
)

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("shivamgoel97/largefinetuneparagraph")
# Run inference
sentences = [
    'What steps should be taken if a Sales Representative or customer indicates that the Social Security Number (SSN) on the application is incorrect?',
    'Context: the application if the customer placed an: Internet order using a third-party credit card and the order was updated with the original credit card owner\'s name. Order with the incorrect address on the credit application.When completing an HD/SL or HD/SY application, verify the customer\'s full SSN and refer to the following scenarios: Note: No signature is required if the customer has a valid SSN card. An employer issued W2 can be sent as proof if the customer does not have the SSN card. The address on the driver\'s license does not have to match the address entered on the application.Table: <table border="1" cellpadding="2" cellspacing="1" width="100%"><tbody><tr><th style="text-align: left;" valign="top" width="25%">If\xa0</th><th style="text-align: left;" valign="top">Then\xa0</th></tr><tr><td style="text-align: left;" valign="top" width="25%">a Sales Representative or the customer says the Social Security Number (SSN) on the application is incorrect<br/></td><td style="text-align: left;" valign="top"><div>follow these steps:<br/>\xa0</div><ol><li>Correct the SSN.</li><li>Search the billing system with the SSN to ensure that no accounts appear.</li><li>Resend the application to Known Fraud Bad Debt (KFBD) for scoring.<br/><br/></li></ol>Note: Sales Representatives are not required to rerun the order after receiving confirmation of an incorrect SSN.<br/></td></tr><tr><td style="text-align: left;" valign="top" width="25%">a Sales Representative or the customer says the SSN is correct<br/></td><td style="text-align: left;" valign="top"><p>refer to these scenarios:</p><ul><li>If an application is flagged with "SSN related reason code at Fraud IQ", view the credit report from the vzCredit menu to check if a "V003 - Deceased indicators are present for this consumer, Credit Header Data" message is displayed.<ul><li>If the deceased indicator is found, the application must remain in the <b>HD/SL</b> status and the customer is directed to the Credit Bureau to confirm the customer\'s SSN.</li><li>The deceased indicator is only applicable to the <b>HD/SL</b> hold code.\xa0</li></ul></li><li>If speaking with a Sales Agent, confirm the agent has viewed the customer\'s driver\'s license and SSN card in the store location.<ul><li>If the agent has verified the customer\'s IDs, submit the application to the Credit Bureau for scoring.</li></ul></li><li>If speaking with Inside Sales, CS, directly with a customer or if sales cannot physically verify the documents in-store, complete the following steps:<ol><li>Review the previous decision on the credit application to verify that the application fallout was not due to <b>HD/SN\xa0</b>(if the application is on hold).<ul><li>See<span class="prices"> </span><a href="/content/km/categories/channel-operations/221109.html" style="font-size: 1.0em;font-weight: bold;" target="_blank">IDReveal and HD/SN Applications</a>\xa0if\xa0previously held for <b>HD/SN</b>.<br/></li></ul></li><li>Advise the agent that Verizon requires a copy of the customer\'s Driver\'s License / SSN faxed or emailed to <a href="mailto:[email protected]" target="_blank">[email protected]</a>.<a href="mailto:[email protected]" target="_blank"></a></li><li>Do not request a letter from the Social Security Administration (these are no longer issued).</li></ol></li></ul><p>Note:</p><ul><li>Inside Sales does have the option to send a Digital Documentation Imaging and Authentication (DDIA) link to the customer\'s email to send in documentation for <b>HD/SL</b>.</li><li>Customers are required to send a copy of a driver\'s license and SSN card.</li></ul></td></tr><tr><td style="text-align: left;" valign="top" width="25%">documents are received and the SSN matches the information entered in vzCredit<br/></td><td style="text-align: left;" valign="top"><p>refer to these scenarios:</p><ul><li>If the application has a Fraud IQ alert, resend to the Credit Bureau.</li><li>If the application has an SSN mismatch alert, manually approve the application with the code <b>AP/CA</b>\xa0and the account and line-level class provided by the bureau.\xa0<ul><li>If there is no account and line-level class already provided by the bureau, use:<ul><li>Account level class <b>W</b></li><li>Line level credit class <b>0C\xa0</b></li></ul></li></ul></li><li>Remark the application indicating proof of SSN and Driver\'s License was received.</li></ul></td></tr><tr><td style="text-align: left;" valign="top" width="25%">documents are received and the SSN does not match the information entered in vzCredit<br/></td><td style="text-align: left;" valign="top"><div>follow these steps:<br/>\xa0</div><ol><li>Correct the SSN.</li><li>Resend the application to KFBD for scoring.</li><li>Approve the application with the code\xa0<b>AP</b>/<b>CA.</b></li><li>Approve the application with the following codes if the application does not automatically approve:<ul><li>Account level class<b>\xa0W\xa0</b>\xa0</li><li>Line level credit class<b>\xa00C</b></li></ul></li><li>Remark the application indicating proof\xa0of\xa0SSN and Driver\'s License was received.</li></ol></td></tr></tbody></table>',
    'Customer heard reports that Disney+ accounts were hacked Disney has assured Verizon and the public that there is no indication of a security breach on Disney+. If customers are unable to access their Disney+ account, please follow BAU Support guidelines. If the customer can\'t remember their password on the second screen after logging in, it asks for a password. The customer can use the Forgot Password? link here if they do not remember their password. After successfully logging in, a customer can click on the profile icon in the top right-hand corner; Default is Mickey Mouse, but users can select from one of many different Disney characters here. Under the drop down, choose Account. On the Disney+ account details page the customer can select Change Password link. Customer redeemed Verizon promotion but is being billed for Disney+ by another party Customers that have an existing Disney+ subscription with Disney or a 3rd party may need to take additional steps. Direct Disney+ Annual/Multi Year - No action; Disney will pause and resume their subscription after our promotional period ends. Direct Disney+ Monthly - Existing subscription is cancelled and replaced with the Verizon billed. promotion; Disney will send email advising of next steps 3rd Party billed - Customer must cancel subscription with the 3rd party provider they signed up with; Disney will send email advising of next steps. Please see Existing Disney+ Subscribers for additional details. Disney will send these customers an email advising of the dual subscription. Customers can be directed to Disney’s Help Center for more details. They should select Manage my Account and look for the link for "How do I manage my subscription through Google, Apple or other third-parties?" This Disney webpage provides exact instructions for managing subscriptions through Apple, Google Play, Roku and Amazon. After cancellation a customer can follow these steps to check their Account Details section to confirm they are subscribed through Verizon. Click on the profile icon in the top right-hand corner; Default is Mickey Mouse, but users can select from one of many different Disney characters here. Under the drop down, choose Account. A Disney+ account details page for a customer that has the Verizon promotion active should look like this:',
]
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]

Training Details

Training Dataset

Unnamed Dataset

Size: 30,000 training samples
Columns: sentence_0, sentence_1, and label
Approximate statistics based on the first 1000 samples:
sentence_0 sentence_1 label
type string string int
details
min: 8 tokens
mean: 18.68 tokens
max: 41 tokens

min: 53 tokens
mean: 122.31 tokens
max: 128 tokens

0: ~35.00%
1: ~65.00%

	sentence_0	sentence_1	label
type	string	string	int
details	min: 8 tokens mean: 18.68 tokens max: 41 tokens	min: 53 tokens mean: 122.31 tokens max: 128 tokens	0: ~35.00% 1: ~65.00%

Samples:

sentence_0	sentence_1	label
`What is required to activate a Samsung Galaxy Watch on NumberShare?`	Wearables The Add a Line (AAL) process includes adding Wearables (e.g., Samsung and Apple watches) to an existing account. When adding a Samsung Galaxy Watch: The watch can be activated on NumberShare or as a standalone device. With NumberShare, a Samsung Galaxy Smartwatch is compatible with standard monthly: Android smartphones running OS 5.0 or later Apple iPhone 5 and newer running iOS 9.0 or higher The watch requires a QR Code (emailed after submitting activation of new line) and the Galaxy Wearable app. Refer to Adding a Samsung Watch Process. When adding an Apple Watch 3, 4 or 5: NumberShare is required to activate service on an Apple Watch. To do an AAL for an Apple watch, customers must have a compatible iPhone active on their account or purchase a compatible iPhone with their Apple Watch order. The MDN assigned to the Apple watch in billing is not used for calling or messaging. The Apple Watch shares the MDN of the iPhone/Host. Refer to Adding an Apple Watch Process. Telematic...	`1`
`What type of account creation does not require SPOC approval if requested from specific mailboxes?`	Follow the Verification Matrix to confirm authorization for requests to create a new account: The Verification Matrix is located in Account Verification - Business and Government > Transaction Matrix > Sales - All Channels tab. Creating a Hybrid account does not require SPOC approval if it comes from one of the following Contract Management Mailboxes: Verizon Contract Management - Markets WBG Wireless Contract WBG Wireless Contract Management - Vertical Verizon Contract Management - West Markets (S/L Government - western half of the US) Verizon Contract Management - East Markets (S/L Government - eastern half of the US) Note: If a Hybrid account request comes from one of the above mailboxes, then complete the request. Hybrid Account Creation Refresher GEA/Fulfillment.	`1`
`What system access is required for representatives to get started?`	`To get started, representatives must have access/permissions to customer accounts, which include: ACSS At Your Service (AYS) for system access Familiarity with My Verizon Note: This guide links only to related information on the LNP - Escalations OST page for performing your job. If viewing the OST page in its entirety, be aware it contains additional information for Tier 2 and Tier 3 roles which are separate from the tasks you perform.`	`1`

Loss: CosineSimilarityLoss with these parameters:

{
    "loss_fct": "torch.nn.modules.loss.MSELoss"
}

Training Hyperparameters

Non-Default Hyperparameters

num_train_epochs: 5
multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: no
prediction_loss_only: True
per_device_train_batch_size: 8
per_device_eval_batch_size: 8
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: 5e-05
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1
num_train_epochs: 5
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.0
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: 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
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: batch_sampler
multi_dataset_batch_sampler: round_robin

Training Logs

Epoch	Step	Training Loss
0.1333	500	0.1808
0.2667	1000	0.1659
0.4	1500	0.1628
0.5333	2000	0.1623
0.6667	2500	0.1602
0.8	3000	0.1555
0.9333	3500	0.1545
1.0667	4000	0.1436
1.2	4500	0.1391
1.3333	5000	0.1373
1.4667	5500	0.1355
1.6	6000	0.1371
1.7333	6500	0.1339
1.8667	7000	0.1352
2.0	7500	0.1328
2.1333	8000	0.1176
2.2667	8500	0.119
2.4	9000	0.119
2.5333	9500	0.1192
2.6667	10000	0.1177
2.8	10500	0.1179
2.9333	11000	0.1118
3.0667	11500	0.1086
3.2	12000	0.1088
3.3333	12500	0.1029
3.4667	13000	0.1008
3.6	13500	0.1059
3.7333	14000	0.1065
3.8667	14500	0.1022
4.0	15000	0.1046
4.1333	15500	0.0952
4.2667	16000	0.0939
4.4	16500	0.0979
4.5333	17000	0.0928
4.6667	17500	0.0971
4.8	18000	0.0962
4.9333	18500	0.094

Framework Versions

Python: 3.10.12
Sentence Transformers: 3.3.1
Transformers: 4.47.1
PyTorch: 2.5.1+cu121
Accelerate: 1.2.1
Datasets: 3.2.0
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",
}

shivamgoel97
/

largefinetuneparagraph