SentenceTransformer based on nomic-ai/modernbert-embed-base

This is a sentence-transformers model finetuned from nomic-ai/modernbert-embed-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: nomic-ai/modernbert-embed-base
Maximum Sequence Length: 8192 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': 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("sentence_transformers_model_id")
# Run inference
sentences = [
    'What is the primary purpose of the FITS format?',
    'FITS format\n\nAll of the ODF/SDF component files, with the exception of the summary\nfiles, reconstructed orbit file, and raw attitude file, are FITS files\nand conform to the standard. A description of the FITS format can be\nfound in , which is accessible also at the URL\nThe calibration files and the bulk of the PPS products also conform to\nthe FITS standard. Wherever possible and desirable the calibration files\nand the PPS products follow the conventions of the OGIP\n(http://heasarc.gsfc.nasa.gov/docs/heasarc/ofwg/ofwg_intro.html) (Office\nof Guest Investigator Programs) FITS working group. The HEASARC FITS\nWorking Group activities are described at the following URL:\nFor FITS files where OGIP FITS standards are not applicable or\navailable, new standards closely following the OGIP approach are used.\n\nThe FITS format is primarily designed to store scientific data sets\nconsisting of multidimensional arrays (1-D spectra, 2-D images or 3-D\ndata cubes) and 2-dimensional tables containing rows and columns of\ndata. A FITS data file is composed of a sequence of Header + Data Units\n(HDUs).\n\nThe general structure of a FITS file is as follows:\n\n-   a primary header;\n\n-   a primary data array of zero length;\n\n-   zero or more extensions\n\nEach extension consists of an extension header and a data section.\nExtensions are named and can appear in any order. Only the following\nFITS extensions are used:\n\n-   ASCII table: XTENSION=TABLE\n\n-   binary table: XTENSION=BINTABLE\n\n-   image: XTENSION=IMAGE\n\nThe header consists of keyword=value statements, which describe the\norganisation of the data in the HDU and the format of the contents. It\nmay also provide additional information, for example, about instrument\nstatus or the history of the data. The following block contains the\ndata, which are structured as specified in the header. The data section\nof the HDU may contain a digital image, a table or a multidimensional\nmatrix that is not an image. An HDU need not contain data.\n',
    'ASCII\n\nASCII files are used to present script and some tabular information. In\nparticular, each ODF/SDF contains a single summary file, with a summary\nof the information relating to the observation or slew (see\nSect.\xa0[dfhb:par:odf]).\n',
]
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: 3,625 training samples
Columns: anchor and positive
Approximate statistics based on the first 1000 samples:
anchor positive
type string string
details
min: 2 tokens
mean: 15.71 tokens
max: 38 tokens

min: 2 tokens
mean: 412.57 tokens
max: 3755 tokens

	anchor	positive
type	string	string
details	min: 2 tokens mean: 15.71 tokens max: 38 tokens	min: 2 tokens mean: 412.57 tokens max: 3755 tokens

Samples:

anchor	positive
`What is the purpose of the document described in the preface?`	Preface This is the reference document describing the individual XMM-Newton Survey Science Centre (SSC) data product files. It is intended to be of use to software developers, archive administrators and to scientists analysing XMM-Newton data. Please see the SSC data products Interface Control Document (XMM-SOC-ICD-0006-SSC, issue 4.0) for a description of the product group files and other related files that are sent to the SOC. This version (4.3) includes changes related to the upgrade to SAS16.0 in the processing pipeline originally developped in 2012 to uniformly process all the XMM data at that time, from which the 3XMM catalogue was derived. Revisions and additions since version 4.2 are identified by change bars at the right of each page. This document will continue to evolve through subsequent issues, under indirect control from the SAS and SSC configuration control boards. This document is the result of the work of many people. Contributors have included: Hermann Brunner, G...
`What version of the document is described in the preface?`	Preface This is the reference document describing the individual XMM-Newton Survey Science Centre (SSC) data product files. It is intended to be of use to software developers, archive administrators and to scientists analysing XMM-Newton data. Please see the SSC data products Interface Control Document (XMM-SOC-ICD-0006-SSC, issue 4.0) for a description of the product group files and other related files that are sent to the SOC. This version (4.3) includes changes related to the upgrade to SAS16.0 in the processing pipeline originally developped in 2012 to uniformly process all the XMM data at that time, from which the 3XMM catalogue was derived. Revisions and additions since version 4.2 are identified by change bars at the right of each page. This document will continue to evolve through subsequent issues, under indirect control from the SAS and SSC configuration control boards. This document is the result of the work of many people. Contributors have included: Hermann Brunner, G...
`What is the main change in version 4.3 of the document?`	Preface This is the reference document describing the individual XMM-Newton Survey Science Centre (SSC) data product files. It is intended to be of use to software developers, archive administrators and to scientists analysing XMM-Newton data. Please see the SSC data products Interface Control Document (XMM-SOC-ICD-0006-SSC, issue 4.0) for a description of the product group files and other related files that are sent to the SOC. This version (4.3) includes changes related to the upgrade to SAS16.0 in the processing pipeline originally developped in 2012 to uniformly process all the XMM data at that time, from which the 3XMM catalogue was derived. Revisions and additions since version 4.2 are identified by change bars at the right of each page. This document will continue to evolve through subsequent issues, under indirect control from the SAS and SSC configuration control boards. This document is the result of the work of many people. Contributors have included: Hermann Brunner, G...

Loss: CachedMultipleNegativesRankingLoss with these parameters:

{
    "scale": 1.0,
    "similarity_fct": "get_similarity"
}

Evaluation Dataset

Unnamed Dataset

Size: 30 evaluation samples
Columns: anchor and positive
Approximate statistics based on the first 30 samples:
anchor positive
type string string
details
min: 10 tokens
mean: 16.73 tokens
max: 31 tokens

min: 6 tokens
mean: 655.37 tokens
max: 6762 tokens

	anchor	positive
type	string	string
details	min: 10 tokens mean: 16.73 tokens max: 31 tokens	min: 6 tokens mean: 655.37 tokens max: 6762 tokens

Samples:

anchor	positive
`In pn imaging mode event lists, what is the type of the OFFSETX column?`	- In pn event lists this extension contains the CCD columns to which an additional offset is applied to reduce noise (the offset is later subtracted again by the SAS). In the MOS event lists this extension currently defines columns outside the sensitive CCD window, to which formal very high values of the offset are associated. These columns are discarded by the data processing. - For MOS imaging mode event lists this extension contains the following columns: Name Type Description --------- ---------------- ---------------------------------------------------------- RAWX 2-byte INTEGER Row or column of the bad offset OFFSETX 2-byte INTEGER amplitude of additional column offset (0 for row offset) OFFSETY 2-byte INTEGER amplitude of additional row offset (0 for column offset) CCDNR 1-byte CCD where the offset occurs - For MOS timing mode event lists this extension contains the...
`What are the three binary table extensions created per source used for?`	- This product lists bright sources detected by EPIC which fall in the RGS field of view. It also includes the entries for the proposal position and the on-axis location. EPIC and RGS positions are given, as well as RGS spatial and energy-dispersion angle extraction regions for the sources and a background region. - These files are identified using the keyword CONTENT = 'RGS SOURCE LIST' / File content in the primary header. - There are two binary table extensions (SRCLIST and RGSn_BACKGROUND), plus a further three binary table extensions per source (RGSn_SRCm_SPATIAL, RGSn_SRCm_ORDER_1 and RGSn_SRCm_ORDER_2, where n is the number of the RGS (1 or 2) and m is the number of the source. - The SRCLIST extension has the following columns: Name Type Description -------------- ------------------ --------------------------------------------------------- INDEX 2-byte INTEGER Source inde...
`What is the purpose of the analysis steps outlined in the document?`	Structure of the document The structure of the present document is as follows: - Chapter [sasguide:par:analysis] introduces the investigator to the analysis of XMM-Newton (http://www.cosmos.esa.int/web/xmm-newton/technical-details) data. It provides a brief description of XMM-Newton (http://www.cosmos.esa.int/web/xmm-newton/technical-details) observation and calibration files and outlines the analysis steps required to produce calibrated event files and to extract scientific products. - Chapter [sasguide:par:gui] describes the SAS graphical user interface (GUI), a user friendly tool which enables SAS interactive analysis tasks to be run without using the command line. - Chapters [sasguide:par:epic], [sasguide:par:rgs] and [sasguide:par:om] describe the SAS analysis steps required to obtain EPIC (http://www.cosmos.esa.int/web/xmm-newton/technical-details-epic), RGS (http://www.cosmos.esa.int/web/xmm-newton/technical-details-r...

Loss: CachedMultipleNegativesRankingLoss with these parameters:

{
    "scale": 1.0,
    "similarity_fct": "get_similarity"
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 16
per_device_eval_batch_size: 4
num_train_epochs: 2
lr_scheduler_type: constant
warmup_ratio: 0.1
bf16: True
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 16
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: 1.0
num_train_epochs: 2
max_steps: -1
lr_scheduler_type: constant
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: 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
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	Validation Loss
0.0441	10	2.3929	-
0.0881	20	2.2876	-
0.1322	30	2.2502	-
0.1762	40	2.2265	-
0.2203	50	2.176	0.9569
0.2643	60	2.1931	-
0.3084	70	2.1666	-
0.3524	80	2.1637	-
0.3965	90	2.1684	-
0.4405	100	2.1373	0.9265
0.4846	110	2.135	-
0.5286	120	2.1159	-
0.5727	130	2.113	-
0.6167	140	2.098	-
0.6608	150	2.0931	0.9054
0.7048	160	2.0954	-
0.7489	170	2.0882	-
0.7930	180	2.0926	-
0.8370	190	2.1139	-
0.8811	200	2.1151	0.8745
0.9251	210	2.1033	-
0.9692	220	2.1014	-
1.0132	230	2.0139	-
1.0573	240	2.0408	-
1.1013	250	2.0257	0.9039
1.1454	260	2.0401	-
1.1894	270	2.0189	-
1.2335	280	2.0521	-
1.2775	290	2.055	-
1.3216	300	2.0407	0.9321
1.3656	310	2.0252	-
1.4097	320	2.0126	-
1.4537	330	2.0431	-
1.4978	340	2.0293	-
1.5419	350	2.042	0.9105
1.5859	360	2.0557	-
1.6300	370	2.0481	-
1.6740	380	2.0169	-
1.7181	390	2.0402	-
1.7621	400	2.0376	0.8873
1.8062	410	2.045	-
1.8502	420	1.9934	-
1.8943	430	2.0335	-
1.9383	440	2.0278	-
1.9824	450	2.0313	0.8658

Framework Versions

Python: 3.10.14
Sentence Transformers: 3.4.1
Transformers: 4.49.0
PyTorch: 2.6.0+cu124
Accelerate: 1.4.0
Datasets: 3.3.2
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",
}

CachedMultipleNegativesRankingLoss

@misc{gao2021scaling,
    title={Scaling Deep Contrastive Learning Batch Size under Memory Limited Setup},
    author={Luyu Gao and Yunyi Zhang and Jiawei Han and Jamie Callan},
    year={2021},
    eprint={2101.06983},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}