ModernBERT Embed base Legal Matryoshka

This is a sentence-transformers model finetuned from nomic-ai/nomic-embed-text-v2-moe on the json dataset. 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

Model Description

Model Type: Sentence Transformer
Base model: nomic-ai/nomic-embed-text-v2-moe
Maximum Sequence Length: 512 tokens
Output Dimensionality: 768 dimensions
Similarity Function: Cosine Similarity
Training Dataset:
- json
Language: en
License: apache-2.0

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': 512, 'do_lower_case': False}) with Transformer model: NomicBertModel 
  (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})
  (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("tsss1/expressvpn_embeddingmodel")
# Run inference
sentences = [
    'Last updated: October 21, 2024\n\nThis guide is for users who are having issues streaming Max (formerly HBO Max) while connected to the VPN.\n\nTo comply with the Max Terms of Use and ExpressVPN Terms of Service, you should connect to a server location that matches the country where you are currently located.\n\nJump to…\n\n1. Change to a different VPN server location\n2. Sign out of the Max app, then sign in again\n3. Watch Max using your browser\n4. Contact ExpressVPN Support\n\n1. Change to a different VPN server location\n\nIf you are a U.S. user having issues streaming Max, try changing to these VPN server locations in the following order:\n\nUSA – San Francisco\nUSA – Washington DC\nUSA – New York\nUSA – Los Angeles – 1\n\nBelow are instructions for changing your VPN server location on:\n\nWindows\nMac\niOS\nAndroid\nAndroid TV\nApple TV\nLinux\nRouters\nIf you are streaming via the Max app, you should force-close it and reopen it each time you change location. Below are instructions for force-closing an app on:iOS: Swipe up from the bottom of the homescreen, keeping your finger pressed until app previews appear at left. Swipe to find the Max app preview, then swipe up to close the app.\n\nAndroid: On your Android device, open your multitasking interface. The way to do this varies depending on your device:\n\nIf your device has three icons at the bottom of the screen, tap either the three vertical lines icon or the square icon.\nIf your device features a single horizontal line at the bottom of the screen, swipe up from the bottom to the middle of the screen, hold for a second, then release.\n\nNext, swipe to find the Max app preview, then swipe to force-close the app. The direction you need to swipe will vary depending on your device.\n\nAndroid TV: Go to Settings, select Apps, and scroll to find the Max app. Select the app, then select Force Stop.\n\nFire TV/Fire Stick: Go to Settings, select Applications, select Manage Installed Applications. Scroll to find the Max app. Select the app, then select Force Stop.\n\nApple TV: Double-click the TV icon on your remote to see the apps currently running. Swipe to find the Max app preview, then swipe up to close the app.\n\nIf you are a non-U.S. user having issues streaming Max, proceed to the next step.\n\nNeed help?\xa0Contact the ExpressVPN Support Team for immediate assistance.\n\nBack to top\n\n2. Sign out of the Max app, then sign in again\n\nIf you are using the Max app, sign out of it, restart your device, and then sign back in.\n\nNeed help?\xa0Contact the ExpressVPN Support Team for immediate assistance.\n\nBack to top\n\n3. Watch Max on your browser\n\nTry streaming Max via your browser by going to https://www.max.com/login and signing in with your Max account details.\n\nIf you are having issues streaming Max from your browser while connected to the VPN:\n\nGet the ExpressVPN browser extension (available for Windows, Mac, and Linux). To use the browser extension, you must also have the ExpressVPN app installed on your computer.\nU.S. users should try connecting to these server locations in the following order:\nUSA – San Francisco\nUSA – Washington DC\nUSA – New York\nUSA – Los Angeles – 1\n\nNon-U.S. users should proceed to the next step.\n\nTry using a different browser. The ExpressVPN browser extension is available on Windows, Mac, and Linux, and it works with Chrome, Firefox, Vivaldi, Chromium, Brave, and Microsoft Edge. The ExpressVPN app must also be installed.\n\nNeed help?\xa0Contact the ExpressVPN Support Team for immediate assistance.\n\nBack to top\n\n4. Contact Support\n\nIf you are still unable to stream Max while connected to the VPN, contact the ExpressVPN Support Team.\n\nBack to top\n\nExpressVPN is optimized to work with Max so you can enjoy online privacy and security all the time, without the VPN interfering. It should never be used as a means of copyright circumvention, which is strictly against our Terms of Service. As we cannot see or control what you do while connected to our VPN, you are responsible at all times for complying with our terms, the Max Terms of Use, and any applicable laws. Compliance requires you to be located in the U.S. while streaming Max with ExpressVPN.\nWas this article helpful?\nYes No',
    'Troubleshooting steps for streaming Max',
    'I can help you with various questions and issues related to ExpressVPN. What do you need assistance with: \n\n',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

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

Evaluation

Metrics

Information Retrieval

Datasets: dim_768, dim_512, dim_256, dim_128 and dim_64
Evaluated with InformationRetrievalEvaluator

Metric	dim_768	dim_512	dim_256	dim_128	dim_64
cosine_accuracy@1	0.6	0.6	0.625	0.625	0.625
cosine_accuracy@3	0.775	0.775	0.775	0.725	0.7
cosine_accuracy@5	0.825	0.875	0.85	0.775	0.75
cosine_accuracy@10	0.875	0.9	0.875	0.875	0.8
cosine_precision@1	0.6	0.6	0.625	0.625	0.625
cosine_precision@3	0.2667	0.2667	0.2667	0.25	0.2417
cosine_precision@5	0.17	0.18	0.175	0.16	0.155
cosine_precision@10	0.09	0.0925	0.09	0.09	0.0825
cosine_recall@1	0.5875	0.5875	0.6125	0.6125	0.6125
cosine_recall@3	0.775	0.775	0.775	0.725	0.7
cosine_recall@5	0.825	0.875	0.85	0.775	0.75
cosine_recall@10	0.875	0.9	0.875	0.875	0.8
cosine_ndcg@10	0.7465	0.7552	0.751	0.7407	0.7078
cosine_mrr@10	0.7042	0.7083	0.7108	0.6992	0.6793
cosine_map@100	0.7103	0.7124	0.7161	0.7051	0.6907

Training Details

Training Dataset

json

Dataset: json
Size: 357 training samples
Columns: positive and anchor
Approximate statistics based on the first 357 samples:
positive anchor
type string string
details
min: 21 tokens
mean: 322.47 tokens
max: 512 tokens

min: 6 tokens
mean: 32.25 tokens
max: 512 tokens

	positive	anchor
type	string	string
details	min: 21 tokens mean: 322.47 tokens max: 512 tokens	min: 6 tokens mean: 32.25 tokens max: 512 tokens

Samples:

positive	anchor
I'd like to discuss common issues that users face when using ExpressVPN. 1. Slow speeds and connectivity issues. 2. Difficulty in setting up ExpressVPN on various devices such as routers, smart TVs, and gaming consoles. 3. Issues with unblocking geo-restricted content on popular streaming services like Netflix, Hulu, and BBC iPlayer. 4. Troubleshooting failed connections and unable to connect to a VPN server. Which one of these topics would you like to discuss further, or is there something else you'd like to bring up?	`I'd be happy to help with any questions or concerns you have about ExpressVPN. What would you like to know or discuss?`
`I'd like to provide information about ExpressVPN, but I think it would be more helpful to get some assistance from you. I'd like to know more about the process of setting up ExpressVPN on a router. Could you explain the general steps to follow and any potential issues that users may encounter during the setup process? Additionally, are there any specific router models that are known to be compatible with ExpressVPN?`	`I can help you with any question you have about ExpressVPN. What is it that you need help with?`
`Last updated: January 11, 2023 Important: If your ExpressVPN free trial or subscription was initiated via the iOS App Store, refer to this guide. This guide will explain how to get or extend an ExpressVPN subscription for iOS users who did not get a free trial or subscription via the App Store. Note: Upgrades and renewals are not currently available within the ExpressVPN app for iOS.`	`ExpressVPN iOS free trial or subscription expiring`

Loss: MatryoshkaLoss with these parameters:

{
    "loss": "MultipleNegativesRankingLoss",
    "matryoshka_dims": [
        768,
        512,
        256,
        128,
        64
    ],
    "matryoshka_weights": [
        1,
        1,
        1,
        1,
        1
    ],
    "n_dims_per_step": -1
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: epoch
per_device_train_batch_size: 2
per_device_eval_batch_size: 2
gradient_accumulation_steps: 4
learning_rate: 2e-05
num_train_epochs: 7
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: True
tf32: False
load_best_model_at_end: True
optim: adamw_torch_fused
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: epoch
prediction_loss_only: True
per_device_train_batch_size: 2
per_device_eval_batch_size: 2
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 4
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: 7
max_steps: -1
lr_scheduler_type: cosine
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: True
fp16: False
fp16_opt_level: O1
half_precision_backend: auto
bf16_full_eval: False
fp16_full_eval: False
tf32: False
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: True
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_fused
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	dim_768_cosine_ndcg@10	dim_512_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.2235	10	2.9921	-	-	-	-	-
0.4469	20	0.9824	-	-	-	-	-
0.6704	30	0.6762	-	-	-	-	-
0.8939	40	0.0133	-	-	-	-	-
0.9832	44	-	0.7669	0.7701	-	-	-
0.2235	10	0.0179	-	-	-	-	-
0.4469	20	0.2714	-	-	-	-	-
0.6704	30	0.0104	-	-	-	-	-
0.8939	40	0.0015	-	-	-	-	-
0.9832	44	-	0.7442	0.7594	0.7465	0.7149	0.7046
1.1341	50	0.2207	-	-	-	-	-
1.3575	60	0.48	-	-	-	-	-
1.5810	70	0.003	-	-	-	-	-
1.8045	80	0.2985	-	-	-	-	-
1.9832	88	-	0.7751	0.774	0.7821	0.7746	0.7365
2.0447	90	0.0168	-	-	-	-	-
2.2682	100	0.0698	-	-	-	-	-
2.4916	110	0.0054	-	-	-	-	-
2.7151	120	0.0112	-	-	-	-	-
2.9385	130	0.0031	-	-	-	-	-
2.9832	132	-	0.7569	0.7537	0.7565	0.7588	0.7251
3.1788	140	0.1794	-	-	-	-	-
3.4022	150	0.3266	-	-	-	-	-
3.6257	160	0.0006	-	-	-	-	-
3.8492	170	0.0003	-	-	-	-	-
3.9832	176	-	0.7491	0.7613	0.7526	0.7513	0.7206
4.0894	180	0.2622	-	-	-	-	-
4.3128	190	0.0004	-	-	-	-	-
4.5363	200	0.0392	-	-	-	-	-
4.7598	210	0.3312	-	-	-	-	-
4.9832	220	0.0021	0.7548	0.7527	0.7466	0.7568	0.7101
5.2235	230	0.7593	-	-	-	-	-
5.4469	240	0.0004	-	-	-	-	-
5.6704	250	0.0003	-	-	-	-	-
5.8939	260	0.0154	-	-	-	-	-
5.9832	264	-	0.7498	0.7545	0.7510	0.7407	0.7147
6.1341	270	0.0162	-	-	-	-	-
6.3575	280	0.447	-	-	-	-	-
6.5810	290	0.001	-	-	-	-	-
6.8045	300	0.1628	-	-	-	-	-
6.9832	308	-	0.7465	0.7552	0.7510	0.7407	0.7078

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.11.11
Sentence Transformers: 3.4.1
Transformers: 4.48.3
PyTorch: 2.3.1+cu121
Accelerate: 1.3.0
Datasets: 3.3.2
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",
}

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}
}