You can deploy this model and start earning money today!

Pravallika2001/
modernbert-embed-base-legal-matryoshka-1

Name: modernbert-embed-base-legal-matryoshka-1
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}) 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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("Pravallika2001/modernbert-embed-base-legal-matryoshka-1")
# Run inference
sentences = [
    'this information to represent the client effectively and, if necessary, \nto advise the client to refrain from wrongful conduct. Almost \nwithout exception, clients come to lawyers in order to determine \ntheir rights and what is, in the complex of laws and regulations, \ndeemed to be legal and correct. Based on experience, lawyers know',
    'What may lawyers advise their clients to refrain from?',
    "Does the regulation’s definition of 'permanent' support the Government’s argument?",
]
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	0.5518	0.5564	0.5193	0.4482	0.3338
cosine_accuracy@3	0.6012	0.5966	0.5657	0.4807	0.3632
cosine_accuracy@5	0.6878	0.6723	0.6445	0.5734	0.4529
cosine_accuracy@10	0.7543	0.7434	0.7156	0.6785	0.5348
cosine_precision@1	0.5518	0.5564	0.5193	0.4482	0.3338
cosine_precision@3	0.5193	0.5188	0.4869	0.4209	0.3168
cosine_precision@5	0.3975	0.3913	0.3716	0.3233	0.2519
cosine_precision@10	0.2297	0.2266	0.2184	0.2068	0.1595
cosine_recall@1	0.2026	0.2039	0.1918	0.161	0.1179
cosine_recall@3	0.5188	0.5184	0.4879	0.418	0.3153
cosine_recall@5	0.6425	0.635	0.603	0.5238	0.412
cosine_recall@10	0.7391	0.7294	0.703	0.6646	0.5143
cosine_ndcg@10	0.6521	0.6476	0.6172	0.5579	0.4263
cosine_mrr@10	0.5987	0.5979	0.5639	0.4948	0.3752
cosine_map@100	0.6393	0.6366	0.6046	0.5373	0.4193

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: 15 tokens
mean: 97.03 tokens
max: 157 tokens
min: 7 tokens
mean: 16.65 tokens
max: 37 tokens

Samples:

positive	anchor
`communications was evidence of the defendant’s guilt; that is, what the defendant said in [object Object]those communications was inculpatory. See id. at 645-52, 674-76. But the State had to [object Object]establish that the communications were the handiwork of the defendant. It was in that [object Object]context that temporal proximity came into play: The timing of the communications relative`	`Which pages of the cited document discuss the defendant's communications and their evidentiary value?`
`lawyer having supervisory authority over performance of specific [object Object]legal work by another lawyer. Whether a lawyer has such [object Object]supervisory authority in particular circumstances is a question of [object Object]fact. Partners and lawyers with comparable authority have at least [object Object]indirect responsibility for all work being done by the firm, while a [object Object]partner or manager in charge of a particular matter ordinarily also`	`Who has at least indirect responsibility for all work being done by the firm?`
`cuando el demandado contesta la demanda y niega su [object Object]responsabilidad total, aunque la acepte posteriormente; [object Object]cuando se defiende injustificadamente de la acción que [object Object]se presenta en su contra; cuando no admite [object Object]francamente su responsabilidad limitada o parcial, a [object Object]pesar de creer que la única razón que tiene para [object Object]oponerse a la demanda es que la cuantía es exagerada;`	`¿Cuál es la razón que el demandado cree tener para oponerse a la demanda?`

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
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: 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}
tp_size: 0
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	89.4929	-	-	-	-	-
1.0	12	-	0.6233	0.6056	0.5715	0.5117	0.3814
1.7033	20	40.7733	-	-	-	-	-
2.0	24	-	0.6495	0.6425	0.6064	0.5491	0.4172
2.5275	30	29.6387	-	-	-	-	-
3.0	36	-	0.6512	0.6476	0.6172	0.5554	0.4252
3.3516	40	26.8564	-	-	-	-	-
3.7033	44	-	0.6521	0.6476	0.6172	0.5579	0.4263

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.11.11
Sentence Transformers: 3.4.1
Transformers: 4.50.0
PyTorch: 2.6.0+cu124
Accelerate: 1.5.2
Datasets: 3.4.1
Tokenizers: 0.21.1

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: 15 tokens mean: 97.03 tokens max: 157 tokens	min: 7 tokens mean: 16.65 tokens max: 37 tokens