SentenceTransformer based on microsoft/mpnet-base

This is a sentence-transformers model finetuned from microsoft/mpnet-base. 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: microsoft/mpnet-base
Maximum Sequence Length: 512 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': 512, 'do_lower_case': False, 'architecture': '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})
)

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("haongn/mpnet-mnr-squad2")
# Run inference
sentences = [
    'When did the Fed begin raising Fed funds rate significantly?',
    'The Fed then raised the Fed funds rate significantly between July 2004 and July 2006. This contributed to an increase in 1-year and 5-year adjustable-rate mortgage (ARM) rates, making ARM interest rate resets more expensive for homeowners. This may have also contributed to the deflating of the housing bubble, as asset prices generally move inversely to interest rates, and it became riskier to speculate in housing. U.S. housing and financial assets dramatically declined in value after the housing bubble burst.',
    'There has been an increasing gulf between the Premier League and the Football League. Since its split with the Football League, many established clubs in the Premier League have managed to distance themselves from their counterparts in lower leagues. Owing in large part to the disparity in revenue from television rights between the leagues, many newly promoted teams have found it difficult to avoid relegation in their first season in the Premier League. In every season except 2001–02 and 2011–12, at least one Premier League newcomer has been relegated back to the Football League. In 1997–98 all three promoted clubs were relegated at the end of the season.',
]
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, 0.4914, 0.0760],
#         [0.4914, 1.0000, 0.1440],
#         [0.0760, 0.1440, 1.0000]])

Training Details

Training Dataset

Unnamed Dataset

Size: 130,308 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: 7 tokens
mean: 14.33 tokens
max: 37 tokens

min: 31 tokens
mean: 159.67 tokens
max: 512 tokens

	sentence_0	sentence_1
type	string	string
details	min: 7 tokens mean: 14.33 tokens max: 37 tokens	min: 31 tokens mean: 159.67 tokens max: 512 tokens

Samples:

sentence_0	sentence_1
`What isn't another name for King Edward's Chair?`	King Edward's Chair (or St Edward's Chair), the throne on which English and British sovereigns have been seated at the moment of coronation, is housed within the abbey and has been used at every coronation since 1308. From 1301 to 1996 (except for a short time in 1950 when it was temporarily stolen by Scottish nationalists), the chair also housed the Stone of Scone upon which the kings of Scots are crowned. Although the Stone is now kept in Scotland, in Edinburgh Castle, at future coronations it is intended that the Stone will be returned to St Edward's Chair for use during the coronation ceremony.[citation needed]
`What is the inflamation a result of?`	While asthma is a well recognized condition, there is not one universal agreed upon definition. It is defined by the Global Initiative for Asthma as "a chronic inflammatory disorder of the airways in which many cells and cellular elements play a role. The chronic inflammation is associated with airway hyper-responsiveness that leads to recurrent episodes of wheezing, breathlessness, chest tightness and coughing particularly at night or in the early morning. These episodes are usually associated with widespread but variable airflow obstruction within the lung that is often reversible either spontaneously or with treatment".
`Other than Chomsky who else published The Sound Pattern of English?`	In 1968 Noam Chomsky and Morris Halle published The Sound Pattern of English (SPE), the basis for generative phonology. In this view, phonological representations are sequences of segments made up of distinctive features. These features were an expansion of earlier work by Roman Jakobson, Gunnar Fant, and Morris Halle. The features describe aspects of articulation and perception, are from a universally fixed set, and have the binary values + or −. There are at least two levels of representation: underlying representation and surface phonetic representation. Ordered phonological rules govern how underlying representation is transformed into the actual pronunciation (the so-called surface form). An important consequence of the influence SPE had on phonological theory was the downplaying of the syllable and the emphasis on segments. Furthermore, the generativists folded morphophonology into phonology, which both solved and created problems.

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim",
    "gather_across_devices": false
}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 12
per_device_eval_batch_size: 12
num_train_epochs: 1
batch_sampler: no_duplicates
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: 12
per_device_eval_batch_size: 12
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: 1
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: 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: 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}
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: round_robin
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss
0.0460	500	0.5794
0.0921	1000	0.1259
0.1381	1500	0.1169
0.1842	2000	0.1045
0.2302	2500	0.0858
0.2763	3000	0.0835
0.3223	3500	0.0758
0.3684	4000	0.0785
0.4144	4500	0.0591
0.4604	5000	0.0644
0.5065	5500	0.063
0.5525	6000	0.0537
0.5986	6500	0.0575
0.6446	7000	0.0537
0.6907	7500	0.0473
0.7367	8000	0.0599
0.7828	8500	0.0542
0.8288	9000	0.0464
0.8749	9500	0.05
0.9209	10000	0.0455
0.9669	10500	0.043

Framework Versions

Python: 3.12.11
Sentence Transformers: 5.1.1
Transformers: 4.56.1
PyTorch: 2.8.0+cu126
Accelerate: 1.10.1
Datasets: 4.0.0
Tokenizers: 0.22.0

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",
}

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}
}