This is a sentence-transformers model finetuned from sentence-transformers/all-mpnet-base-v2. 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: sentence-transformers/all-mpnet-base-v2
Maximum Sequence Length: 384 tokens
Output Dimensionality: 768 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': 384, 'do_lower_case': False}) with Transformer model: MPNetModel 
  (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("sentence_transformers_model_id")
# Run inference
sentences = [
    'specifically, the proposed regulations proposed the addition of new sec. 300.13 to the user fee regulations to establish a $67 user fee for issuing an estate tax closing letter for an estate.',
    'additionally, the preamble to the proposed regulations explains the special benefits conferred by the issuance of estate tax closing letters and analyzes how the irs has computed that the full cost of issuing an estate tax closing letter is $67.',
    'with respect to whether and how the partnership allocates the rehabilitation credit to partners, the comment specifically asked ``whether the partners are allocated 20 percent of the credit each year although all of the credit basis is reduced in the first year when the property is placed in service or whether, after the first year, the remaining four years over which the credit is spread is taken into account and applied solely at the partner level over those remaining years, consistent with the section 1.50-1 regulations.',
]
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]

Training Details

Training Dataset

Unnamed Dataset

Size: 24,004 training samples
Columns: sentence_0 and sentence_1
Approximate statistics based on the first 1000 samples:
sentence_0 sentence_1
type string string
details
min: 6 tokens
mean: 56.37 tokens
max: 384 tokens
min: 8 tokens
mean: 59.22 tokens
max: 384 tokens

Samples:

sentence_0	sentence_1
`in such cases, the staff will accept the use of the simplified method for only some but not all share option grants.`	`if a company uses this simplified method, the company should disclose in the notes to its financial statements the use of the method, the reason why the method was used, the types of share option grants for which the method was used if the method was not used for all share option grants, and the periods for which the method was used if the method was not used in all periods.`
background subject to various exceptions, section 6033(a)(1) of the internal revenue code (code) requires every organization exempt from taxation under section 501(a) (tax-exempt organization) to file an annual return, stating specifically the items of gross income, receipts, and disbursements, and such other information for the purpose of carrying out the internal revenue laws as the secretary of the treasury or his delegate (secretary) may by forms or regulations prescribe, and keep such records, render under oath such statements, make such other returns, and comply with such rules and regulations as the secretary may from time to time prescribe.	the annual information returns required under section 6033 are forms 990, [object Object]short form return of organization exempt from income tax;'' 990-pf, [object Object]information and initial excise tax return for black lung benefit trusts and certain related persons.'' annual returns filed by tax- exempt organizations, section 527 organizations, nonexempt private foundations described in section 6033(d), and section 4947(a)(1) trusts (which are both treated as organizations described in section 501(c)(3) for this purpose) are information returns intended to help ensure that the filing organizations comply with applicable federal tax laws.
`interpretive response: no. before becoming a public entity, company a did not use the fair-value-based method for either its share options or its liability awards.`	`\12\ \12\ this view is consistent with the fasb's basis for rejecting full retrospective application of fasb asc topic 718 as described in the basis for conclusions of statement 123r, paragraph b251.`

Loss: MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim" 4}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 4
per_device_eval_batch_size: 4
fp16: True
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: 4
per_device_eval_batch_size: 4
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: 3
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: True
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
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
0.0833	500	0.5984
0.1666	1000	0.419
0.2500	1500	0.3454
0.3333	2000	0.3111
0.4166	2500	0.2628
0.4999	3000	0.2747
0.5832	3500	0.2567
0.6666	4000	0.2184
0.7499	4500	0.1802
0.8332	5000	0.1796
0.9165	5500	0.174
0.9998	6000	0.1742
1.0832	6500	0.1043
1.1665	7000	0.1011
1.2498	7500	0.1193
1.3331	8000	0.1167
1.4164	8500	0.1037
1.4998	9000	0.1097
1.5831	9500	0.1018
1.6664	10000	0.1017
1.7497	10500	0.1028
1.8330	11000	0.0854
1.9163	11500	0.088
1.9997	12000	0.1027
2.0830	12500	0.0778
2.1663	13000	0.0645
2.2496	13500	0.0503
2.3329	14000	0.0822
2.4163	14500	0.0616
2.4996	15000	0.0688
2.5829	15500	0.0543
2.6662	16000	0.0678
2.7495	16500	0.0565
2.8329	17000	0.0683
2.9162	17500	0.0412
2.9995	18000	0.0726

Framework Versions

Python: 3.11.13
Sentence Transformers: 4.1.0
Transformers: 4.52.4
PyTorch: 2.6.0+cu124
Accelerate: 1.8.1
Datasets: 3.6.0
Tokenizers: 0.21.2

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",
}

MultipleNegativesRankingLoss