You can deploy this model and start earning money today!

TonyDo/
modernbert-embed-base-legal-matryoshka-2-duy

Name: modernbert-embed-base-legal-matryoshka-2-duy
Availability: InStock

Views

No views yet

ModernBERT Embed base Legal Matryoshka

This is a sentence-transformers model finetuned from nomic-ai/modernbert-embed-base 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/modernbert-embed-base
Maximum Sequence Length: 8192 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': 8192, 'do_lower_case': False, 'architecture': 'ModernBertModel'})
  (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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("TonyDo/modernbert-embed-base-legal-matryoshka-2-duy")
# Run inference
sentences = [
    'time-and-materials contracts, contrary to Plaintiffs’ position.  See Cross-MJAR at 56.  The \ninclusion of these terms in the statute is not inconsistent with Plaintiffs’ interpretation that “based \non hourly rates” references time-and-materials and labor-hour contracts.  Both “delivery order” \nand “cost” are used throughout FAR Subpart 16.6 to prescribe the requirements for time-and-',
    "What do Plaintiffs argue the phrase 'based on hourly rates' references?",
    "What do SHS and VCH MJAR both state about the Agency's evaluation of protégés?",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.5310, 0.1946],
#         [0.5310, 1.0000, 0.2093],
#         [0.1946, 0.2093, 1.0000]])

Evaluation

Metrics

Information Retrieval

Dataset: dim_768
Evaluated with InformationRetrievalEvaluator with these parameters:
json
1{ 2 "truncate_dim": 768 3}

Metric	Value
cosine_accuracy@1	0.5641
cosine_accuracy@3	0.609
cosine_accuracy@5	0.6847
cosine_accuracy@10	0.7496
cosine_precision@1	0.5641
cosine_precision@3	0.5265
cosine_precision@5	0.4006
cosine_precision@10	0.2304
cosine_recall@1	0.2137
cosine_recall@3	0.527
cosine_recall@5	0.6462
cosine_recall@10	0.7378
cosine_ndcg@10	0.658
cosine_mrr@10	0.6078
cosine_map@100	0.6474

Information Retrieval

Dataset: dim_512
Evaluated with InformationRetrievalEvaluator with these parameters:
json
1{ 2 "truncate_dim": 512 3}

Metric	Value
cosine_accuracy@1	0.5456
cosine_accuracy@3	0.5873
cosine_accuracy@5	0.6723
cosine_accuracy@10	0.7388
cosine_precision@1	0.5456
cosine_precision@3	0.507
cosine_precision@5	0.387
cosine_precision@10	0.227
cosine_recall@1	0.2069
cosine_recall@3	0.5116
cosine_recall@5	0.6274
cosine_recall@10	0.7282
cosine_ndcg@10	0.6435
cosine_mrr@10	0.59
cosine_map@100	0.6316

Information Retrieval

Dataset: dim_256
Evaluated with InformationRetrievalEvaluator with these parameters:
json
1{ 2 "truncate_dim": 256 3}

Metric	Value
cosine_accuracy@1	0.5224
cosine_accuracy@3	0.5611
cosine_accuracy@5	0.6383
cosine_accuracy@10	0.728
cosine_precision@1	0.5224
cosine_precision@3	0.4858
cosine_precision@5	0.3691
cosine_precision@10	0.2201
cosine_recall@1	0.1969
cosine_recall@3	0.4874
cosine_recall@5	0.6006
cosine_recall@10	0.7079
cosine_ndcg@10	0.619
cosine_mrr@10	0.5659
cosine_map@100	0.6076

Information Retrieval

Dataset: dim_128
Evaluated with InformationRetrievalEvaluator with these parameters:
json
1{ 2 "truncate_dim": 128 3}

Metric	Value
cosine_accuracy@1	0.4544
cosine_accuracy@3	0.4915
cosine_accuracy@5	0.5765
cosine_accuracy@10	0.6646
cosine_precision@1	0.4544
cosine_precision@3	0.4292
cosine_precision@5	0.3323
cosine_precision@10	0.2019
cosine_recall@1	0.1676
cosine_recall@3	0.4235
cosine_recall@5	0.5359
cosine_recall@10	0.6453
cosine_ndcg@10	0.5536
cosine_mrr@10	0.4995
cosine_map@100	0.5423

Information Retrieval

Dataset: dim_64
Evaluated with InformationRetrievalEvaluator with these parameters:
json
1{ 2 "truncate_dim": 64 3}

Metric	Value
cosine_accuracy@1	0.3539
cosine_accuracy@3	0.3833
cosine_accuracy@5	0.4544
cosine_accuracy@10	0.5301
cosine_precision@1	0.3539
cosine_precision@3	0.3251
cosine_precision@5	0.2532
cosine_precision@10	0.1592
cosine_recall@1	0.1373
cosine_recall@3	0.3314
cosine_recall@5	0.4159
cosine_recall@10	0.5155
cosine_ndcg@10	0.4376
cosine_mrr@10	0.3907
cosine_map@100	0.4342

Training Details

Training Dataset

json

Dataset: json
Size: 5,822 training samples
Columns: positive and anchor
Approximate statistics based on the first 1000 samples:
positive anchor
type string string
details
min: 29 tokens
mean: 96.94 tokens
max: 170 tokens
min: 8 tokens
mean: 16.55 tokens
max: 41 tokens

Samples:

positive	anchor
`Apelaciones, Mech-Tech College; Mech-Tech Management; LLC, [object Object]Artificial Intelligence, Corp. d/b/a Artificial Intelligence, Corp; [object Object]Compañía Aseguradora ABC; Compañía Aseguradora DEF y [object Object]Compañía Aseguradora XYZ (en adelante, parte apelante) mediante [object Object]Apelación Civil. En la misma, nos solicita que revisemos la Sentencia [object Object]emitida y notificada el 15 de septiembre de 2023, por el Tribunal de`	`What is the name of the college involved in the case?`
`otra parte. Fernández v. San Juan Cement Co., Inc., 118 DPR 713, [object Object]718-719 (1987). Nuestro más Alto Foro ha dispuesto que, la [object Object]facultad de imponer honorarios de abogados es la mejor arma que [object Object] [object Object]22 Id. [object Object]23 Andamios de PR v. Newport Bonding, 179 DPR 503, 520 (2010); Pérez Rodríguez [object Object]v. López Rodríguez, supra; SLG González -Figueroa v. Pacheco Romero, supra;`	`Which case reference number ends with 503?`
`necessary. Some firms, for example, have a procedure whereby [object Object]junior lawyers can make confidential referral of ethical problems [object Object]directly to a designated supervising lawyer or special committee. [object Object]See rule 4-5.2. Firms, whether large or small, may also rely on [object Object]continuing legal education in professional ethics. In any event the [object Object]ethical atmosphere of a firm can influence the conduct of all its`	`What kind of legal education is mentioned in the context of firms' ethical measures?`

Loss: MatryoshkaLoss with these parameters:
json
1{ 2 "loss": "MultipleNegativesRankingLoss", 3 "matryoshka_dims": [ 4 768, 5 512, 6 256, 7 128, 8 64 9 ], 10 "matryoshka_weights": [ 11 1, 12 1, 13 1, 14 1, 15 1 16 ], 17 "n_dims_per_step": -1 18}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: epoch
per_device_train_batch_size: 32
per_device_eval_batch_size: 16
gradient_accumulation_steps: 16
learning_rate: 2e-05
num_train_epochs: 4
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: True
load_best_model_at_end: True
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: 32
per_device_eval_batch_size: 16
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 16
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: 4
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: 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: 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}
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
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: False
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: False
prompts: None
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

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
0.8791	10	5.7211	-	-	-	-	-
1.0	12	-	0.6085	0.5977	0.5641	0.5054	0.3930
1.7033	20	2.5747	-	-	-	-	-
2.0	24	-	0.6470	0.6285	0.6096	0.5485	0.4170
2.5275	30	1.9726	-	-	-	-	-
3.0	36	-	0.6583	0.6441	0.6149	0.5515	0.4378
3.3516	40	1.6232	-	-	-	-	-
4.0	48	-	0.658	0.6435	0.619	0.5536	0.4376

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.12.11
Sentence Transformers: 5.1.0
Transformers: 4.56.1
PyTorch: 2.8.0+cu126
Accelerate: 1.10.1
Datasets: 4.1.0
Tokenizers: 0.22.0

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",
}

MatryoshkaLoss

bibtex

@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

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}
}

	positive	anchor
type	string	string
details	min: 29 tokens mean: 96.94 tokens max: 170 tokens	min: 8 tokens mean: 16.55 tokens max: 41 tokens