metadata

language:
  - en
license: apache-2.0
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:5822
  - loss:MatryoshkaLoss
  - loss:MultipleNegativesRankingLoss
base_model: nomic-ai/modernbert-embed-base
widget:
  - source_sentence: >-
      this information about the two documents withheld in part under the
      deliberative-process 

      128 
       
      privilege in No. 11-444, see First Lutz Decl. Ex. DD at 17, 141, but the
      descriptions of the 

      decisionmaking authority are generic, stating that the withheld
      information is a “recommendation 

      from the [FOIA] analyst to his/her supervisor,” id. at 17, and a
      “recommendation from the
    sentences:
      - >-
        What did the plaintiff assert about the CIA's inaccurate
        representations?
      - >-
        What type of document is mentioned as an exhibit in conjunction with the
        withheld documents?
      - >-
        ¿Qué ámbito jurisdiccional es mencionado en el contexto de derechos
        sobre la propia imagen?
  - source_sentence: |-
      Artificial Intelligence, Corp. y que prestaba servicios mediante 
      contrato para el Departamento de Producción de tal corporación. 
      Adujo que, no se encontraba en la obligación de solicitar 
      autorización a la parte apelada para utilizar su imagen, ya que se le 
      había pagado por la producción de múltiples videos publicitarios 
      para el uso de las empresas.  
      Luego 
      de 
      varias 
      incidencias
    sentences:
      - Who has the burden to provide a sufficient record on appeal?
      - >-
        ¿Para qué departamento prestaba servicios Artificial Intelligence, Corp.
        según el contrato?
      - What section numbers are referenced for further information?
  - source_sentence: >-
      submission by protégé firms.  SHS MJAR at 28–30; VCH MJAR at 28–30
      (same).  This, Plaintiffs 

      contend, violates Section 125.8(e) because it purportedly subjects
      protégés to heightened 

      evaluation criteria as compared to offerors generally and makes it harder
      for mentor-protégé JVs 

      to compete against more experienced firms with larger portfolios of past
      work.  SHS MJAR at 28–
    sentences:
      - >-
        On what date were the plaintiff's petition, complaint, and trial court's
        order filed?
      - What section do Plaintiffs contend is violated?
      - What is the amount of pages the party seeks to withhold?
  - source_sentence: >-
      Beginning with the CIA’s submissions, the CIA states in its declaration
      submitted in No. 

      11-445 that “[s]ome of the records for which information has been withheld
      pursuant to 

      Exemption (b)(5) contain confidential communications between CIA staff and
      attorneys within 

      the CIA’s Office of General Counsel about the processing of certain FOIA
      requests.”  Third Lutz
    sentences:
      - >-
        What is the subject of the confidential communications mentioned in the
        document?
      - >-
        Which rule number is associated with the responsibilities regarding
        nonlawyer assistants?
      - ¿Qué número de referencia tiene el documento?
  - source_sentence: >-
      contracting/contracting-assistance-programs/sba-mentor-protege-program
      (last visited Apr. 19, 

      2023). 

      5 
       
      protégé must demonstrate that the added mentor-protégé relationship will
      not adversely affect the 

      development of either protégé firm (e.g., the second firm may not be a
      competitor of the first 

      firm).”  13 C.F.R. § 125.9(b)(3).
    sentences:
      - >-
        What discretion do district courts have regarding a defendant’s
        invocation of FOIA exemptions?
      - What must the protégé demonstrate about the mentor-protégé relationship?
      - >-
        Which exemptions are mentioned in relation to the plaintiff's
        accusations?
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - cosine_accuracy@1
  - cosine_accuracy@3
  - cosine_accuracy@5
  - cosine_accuracy@10
  - cosine_precision@1
  - cosine_precision@3
  - cosine_precision@5
  - cosine_precision@10
  - cosine_recall@1
  - cosine_recall@3
  - cosine_recall@5
  - cosine_recall@10
  - cosine_ndcg@10
  - cosine_mrr@10
  - cosine_map@100
model-index:
  - name: ModernBERT Embed base Legal Matryoshka
    results:
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 768
          type: dim_768
        metrics:
          - type: cosine_accuracy@1
            value: 0.5285935085007728
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.5718701700154559
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.6646058732612056
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.7310664605873262
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.5285935085007728
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.5141679546625451
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.3941267387944359
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.2329211746522411
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.17877382792375063
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.4894384337970118
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.6120556414219475
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.7184441009788768
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.6300476733345887
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.5741100561811532
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.6186392686743281
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 512
          type: dim_512
        metrics:
          - type: cosine_accuracy@1
            value: 0.5162287480680062
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.5486862442040186
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.6414219474497682
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.7171561051004637
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.5162287480680062
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.4981968057702215
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.38083462132921175
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.22720247295208656
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.17400824317362185
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.47346728490468826
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.5910613086038125
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.702344152498712
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.6137901932050573
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.5592913569343243
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.6021884440021203
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 256
          type: dim_256
        metrics:
          - type: cosine_accuracy@1
            value: 0.482225656877898
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.5285935085007728
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.598145285935085
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.678516228748068
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.482225656877898
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.46986089644513135
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.35857805255023184
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.21468315301391033
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.16267387944358577
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.4492529623905203
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.5569294178258629
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.6642194744976816
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.5781404945062661
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.5249122936139936
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.5698418441661705
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 128
          type: dim_128
        metrics:
          - type: cosine_accuracy@1
            value: 0.41576506955177744
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.4435857805255023
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.5363214837712519
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.6105100463678517
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.41576506955177744
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.3992787223080887
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.31282843894899537
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.19242658423493045
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.14258114374034003
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.3835651725914477
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.48776403915507466
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.5963420917053066
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.5108672198469205
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.4573213365717227
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.5029873598412773
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 64
          type: dim_64
        metrics:
          - type: cosine_accuracy@1
            value: 0.312210200927357
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.3508500772797527
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.43585780525502316
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.47913446676970634
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.312210200927357
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.3091190108191654
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.250386398763524
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.14976816074188565
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.10497166409067489
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.2954662545079856
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.3930963420917053
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.46805770221535287
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.39563928784117025
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.3508985304580356
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.3939277813526489
            name: Cosine Map@100
datasets:
  - AdamLucek/legal-rag-positives-synthetic

ModernBERT Embed base Legal Matryoshka

This is a sentence-transformers model finetuned from nomic-ai/modernbert-embed-base on the AdamLucek/legal-rag-positives-synthetic 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:
- AdamLucek/legal-rag-positives-synthetic
Language: en
License: apache-2.0

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 8192, 'do_lower_case': False}) with Transformer model: 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.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("AdamLucek/ModernBERT-embed-base-legal-MRL")
# Run inference
sentences = [
    'contracting/contracting-assistance-programs/sba-mentor-protege-program (last visited Apr. 19, \n2023). \n5 \n \nprotégé must demonstrate that the added mentor-protégé relationship will not adversely affect the \ndevelopment of either protégé firm (e.g., the second firm may not be a competitor of the first \nfirm).”  13 C.F.R. § 125.9(b)(3).',
    'What must the protégé demonstrate about the mentor-protégé relationship?',
    'What discretion do district courts have regarding a defendant’s invocation of FOIA exemptions?',
]
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	0.5286	0.5162	0.4822	0.4158	0.3122
cosine_accuracy@3	0.5719	0.5487	0.5286	0.4436	0.3509
cosine_accuracy@5	0.6646	0.6414	0.5981	0.5363	0.4359
cosine_accuracy@10	0.7311	0.7172	0.6785	0.6105	0.4791
cosine_precision@1	0.5286	0.5162	0.4822	0.4158	0.3122
cosine_precision@3	0.5142	0.4982	0.4699	0.3993	0.3091
cosine_precision@5	0.3941	0.3808	0.3586	0.3128	0.2504
cosine_precision@10	0.2329	0.2272	0.2147	0.1924	0.1498
cosine_recall@1	0.1788	0.174	0.1627	0.1426	0.105
cosine_recall@3	0.4894	0.4735	0.4493	0.3836	0.2955
cosine_recall@5	0.6121	0.5911	0.5569	0.4878	0.3931
cosine_recall@10	0.7184	0.7023	0.6642	0.5963	0.4681
cosine_ndcg@10	0.63	0.6138	0.5781	0.5109	0.3956
cosine_mrr@10	0.5741	0.5593	0.5249	0.4573	0.3509
cosine_map@100	0.6186	0.6022	0.5698	0.503	0.3939

Training Details

AdamLucek/legal-rag-positives-synthetic

Dataset: AdamLucek/legal-rag-positives-synthetic
Size: 5,822 training samples
Columns: positive and anchor
Approximate statistics based on the first 1000 samples:
positive anchor
type string string
details
min: 15 tokens
mean: 97.6 tokens
max: 153 tokens

min: 8 tokens
mean: 16.68 tokens
max: 41 tokens

	positive	anchor
type	string	string
details	min: 15 tokens mean: 97.6 tokens max: 153 tokens	min: 8 tokens mean: 16.68 tokens max: 41 tokens

Samples:

positive	anchor
`infrastructure security information,” the information at issue must, “if disclosed . . . reveal vulner- abilities in Department of Defense critical infrastructure.” 10 U.S.C. § 130e(f). The closest the Department comes is asserting that the information “individually or in the aggregate, would enable`	`What type of information must reveal vulnerabilities if disclosed?`
`they have bid.” Oral Arg. Tr. at 42:18–20. Plaintiffs also assert that, should this Court require the Polaris Solicitations to consider price at the IDIQ level, such an adjustment “adds a solicitation requirement that would necessarily change the overall structure of the evaluation” GSA must perform in awarding the IDIQ contracts. Oral Arg. Tr. at 43:3–5; see supra Discussion Section`	`Where in the document can further discussion about the assertion be found?`
`otra parte. Fernández v. San Juan Cement Co., Inc., 118 DPR 713, 718-719 (1987). Nuestro más Alto Foro ha dispuesto que, la facultad de imponer honorarios de abogados es la mejor arma que 22 Id. 23 Andamios de PR v. Newport Bonding, 179 DPR 503, 520 (2010); Pérez Rodríguez v. López Rodríguez, supra; SLG González -Figueroa v. Pacheco Romero, supra;`	`What case is cited with the reference number 118 DPR 713?`

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: 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
tf32: True
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: 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: True
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
0.8791	10	5.6528	-	-	-	-	-
1.0	12	-	0.5926	0.5753	0.5457	0.4687	0.3455
1.7033	20	2.4543	-	-	-	-	-
2.0	24	-	0.6195	0.6066	0.5778	0.4998	0.3828
2.5275	30	1.7455	-	-	-	-	-
3.0	36	-	0.6292	0.6135	0.5765	0.5057	0.3928
3.3516	40	1.5499	-	-	-	-	-
3.7033	44	-	0.63	0.6138	0.5781	0.5109	0.3956

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.11.11
Sentence Transformers: 3.3.1
Transformers: 4.48.0
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",
}

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}
}