SAVRN
Search Contact SAVRN

Open-weight model · Fill mask

SecBERT

by Kun jackaduma/SecBERT

This is the pretrained model presented in SecBERT: A Pretrained Language Model for Cyber Security Text, which is a BERT model trained on cyber security text.

Parameters84M
Context514
Weights672.8 MB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads400.6k

Runs On

What it takes to serve SecBERT (84M parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.

PrecisionWeightsMemory neededCheapest setupPer hourAlso fits
16-bit 0.2 GB 0.2 GB 1x MI300X (192 GB)
Vultr		 $1.85 1x H100 $1.99 · 1x MI325X $2.00
8-bit 0.1 GB 0.1 GB 1x MI300X (192 GB)
Vultr		 $1.85 1x H100 $1.99 · 1x MI325X $2.00
4-bit 0.0 GB 0.1 GB 1x MI300X (192 GB)
Vultr		 $1.85 1x H100 $1.99 · 1x MI325X $2.00

Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Sep 18, 2026.

SAVRN's Notes on SecBERT

Security text has its own vocabulary, and this model was built around that fact: a 52,000-entry wordpiece vocabulary the publisher calls secvocab, matched to a corpus of cyber security papers. The fill-mask head is the starting point, not the product; the publisher positions it as a base for NER, text classification and Q&A. Six layers and 84M parameters come to 0.2 GB at 16-bit, so the cheapest Index setup, one MI300X with 192 GB at $1.85 per hour on-demand, is a card to share across many fine-tunes, not dedicate to one.

Apache 2.0 means a fine-tuned derivative can ship commercially, redistributed, with the notices kept and changes stated. Check the 514-token context first: documents longer than that get chunked. Then compare your text against the training sets, APTnotes, Stucco-Data and CASIE, and note the publisher also trained a SecRoBERTa version and last updated this one on 2023-06-26.

SAVRN Research, 2026-09-18

Model Card

By Kun, published under apache-2.0, revision 7c603df5bc4c.

This is the pretrained model presented in SecBERT: A Pretrained Language Model for Cyber Security Text, which is a BERT model trained on cyber security text. The training corpus was papers taken from SecBERT has its own wordpiece vocabulary (secvocab) that's built to best match the training corpus. We trained SecBERT and SecRoBERTa versions. We proposed to build language model which work on cyber security text, as result, it can improve downstream tasks (NER, Text Classification, Semantic Understand, Q&A) in Cyber Security Domain. First, as below shows Fill-Mask pipeline in Google Bert, AllenAI SciBert and our SecBERT. The original repo can be found here.

Read Kun's full model card

This is the pretrained model presented in SecBERT: A Pretrained Language Model for Cyber Security Text, which is a BERT model trained on cyber security text.

The training corpus was papers taken from * APTnotes * Stucco-Data: Cyber security data sources * CASIE: Extracting Cybersecurity Event Information from Text * SemEval-2018 Task 8: Semantic Extraction from CybersecUrity REports using Natural Language Processing (SecureNLP).

SecBERT has its own wordpiece vocabulary (secvocab) that's built to best match the training corpus.

We trained SecBERT and SecRoBERTa versions.

Available models include: * SecBERT * SecRoBERTa

Fill Mask

We proposed to build language model which work on cyber security text, as result, it can improve downstream tasks (NER, Text Classification, Semantic Understand, Q&A) in Cyber Security Domain.

First, as below shows Fill-Mask pipeline in Google Bert, AllenAI SciBert and our SecBERT .

The original repo can be found here.

Configuration

Architecture
BertForMaskedLM
Context length (tokens)
514
Layers
6
Hidden size
768
Feed-forward size
3,072
Attention heads
12
Vocabulary size
52,000
Model type
bert

Identity and Version

Repository
jackaduma/SecBERT
Publisher
Kun
Task
Fill mask
Modality
Text
Library
transformers
Parameters
84M parameters
Languages
en
Revision
7c603df5bc4c5ba9c731bcc2ea0ab2db36e104cb
First published
2022-03-02
Last updated
2023-06-26

Files and Weights

7 files, 673.3 MB in total. The weights are 2 files totalling 672.8 MB in bin, safetensors.

Weights2 files · 672.8 MB
Configuration1 file · 467 B
Tokenizer1 file · 378.0 KB
Documentation1 file · 1.9 KB
Other1 file · 107.4 KB
Repository1 file · 1.2 KB
Every file
FileTypeSizeSHA-256
model.safetensorsWeights336.4 MB cbc89dbadf58
pytorch_model.binWeights336.4 MB 08a1be6db147
config.jsonConfiguration467 B —
README.mdDocumentation1.9 KB —
fill-mask-result.pngOther107.4 KB —
.gitattributesRepository1.2 KB —
vocab.txtTokenizer378.0 KB —

License and Download

License
apache-2.0
Access
Open weights, no gate
Download size
672.8 MB
Download from Kun

Released by Kun through its official repository on Hugging Face. Read the license.

Built From

  • Trained on (disclosed) APTnotes
  • Trained on (disclosed) CASIE
  • Trained on (disclosed) Stucco-Data

Memory Requirements

PrecisionWeights in memory
As published672.8 MB
16-bit0.2 GB
8-bit0.1 GB
4-bit0.0 GB

Weights only, from the published parameter count; the key-value cache and runtime add to this.

Compare SecBERT

Questions About SecBERT

How much GPU memory does SecBERT need?

About 0.2 GB at 16-bit and 0.1 GB at 4-bit: the weights (84M parameters) plus a working margin. A long context needs more.

What is the cheapest GPU to run SecBERT on?

At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

Can I use SecBERT commercially?

Yes. SecBERT is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.

What is SecBERT's context length?

514 tokens, from the maximum position embeddings in its published configuration.

Similar Models

This model is a distilled version of the RoBERTa-base model. It follows the same training procedure as DistilBERT. The code for the distillation process can be found here. The model has 6 layers, 768 dimension and 12 heads, totalizing 82M parameters (compared to 125M parameters for RoBERTa-base). On average DistilRoBERTa is twice as fast as Roberta-base. We encourage users of this model card to check out the RoBERTa-base model card to learn more about usage, limitations and potential biases. You can use the raw model for masked language modeling, but it's mostly intended to be fine-tuned on a downstream task. See the model hub to look for fine-tuned versions on a task that interests you.…

Open weights apache-2.0 83M parameters 514 tokens transformers
View model

This model is a distilled version of the BERT base model. It was introduced in this paper. The code for the distillation process can be found here. This model is uncased: it does not make a difference between english and English. DistilBERT is a transformers model, smaller and faster than BERT, which was pretrained on the same corpus in a self-supervised fashion, using the BERT base model as a teacher. This means it was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of publicly available data) with an automatic process to generate inputs and labels from those texts using the BERT base model. More precisely, it was pretrained…

Open weights apache-2.0 67M parameters 512 tokens transformers
View model

Model · Fill mask

bert-base-chinese

BERT community

This model has been pre-trained for Chinese, training and random input masking has been applied independently to word pieces (as in the original BERT paper). This model can be used for masked language modeling CONTENT WARNING: Readers should be aware this section contains content that is disturbing, offensive, and can propagate historical and current stereotypes. Significant research has explored bias and fairness issues with language models (see, e.g., Sheng et al. (2021) and Bender et al. (2021)). typevocabsize: 2 vocabsize: 21128 numhiddenlayers: 12

Open weights apache-2.0 103M parameters 512 tokens transformers
View model

Model · Fill mask

bert-base-cased

BERT community

Pretrained model on English language using a masked language modeling (MLM) objective. It was introduced in this paper and first released in this repository. This model is case-sensitive: it makes a difference between english and English. Disclaimer: The team releasing BERT did not write a model card for this model so this model card has been written by the Hugging Face team. BERT is a transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of publicly available data) with an automatic process to generate inputs and labels from…

Open weights apache-2.0 109M parameters 512 tokens transformers
View model

Eval data: Conll03 (NER), GermEval14 (NER), GermEval18 (Classification), GNAD (Classification) Update April 3rd, 2020: we updated the vocabulary file on deepset's s3 to conform with the default tokenization of punctuation tokens. For details see the related FARM issue. If you want to use the old vocab we have also uploaded a "deepset/bert-base-german-cased-oldvocab" model. - We trained using Google's Tensorflow code on a single cloud TPU v2 with standard settings. - We trained 810k steps with a batch size of 1024 for sequence length 128 and 30k steps with sequence length 512. Training took about 9 days. - As training data we used the latest German Wikipedia dump (6GB of raw txt files), the…

Open weights mit 110M parameters 512 tokens transformers
View model

Model · Fill mask

bert-base-uncased

BERT community

Pretrained model on English language using a masked language modeling (MLM) objective. It was introduced in this paper and first released in this repository. This model is uncased: it does not make a difference between english and English. Disclaimer: The team releasing BERT did not write a model card for this model so this model card has been written by the Hugging Face team. BERT is a transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it was pretrained on the raw texts only, with no humans labeling them in any way (which is why it can use lots of publicly available data) with an automatic process to generate inputs and labels from…

Open weights apache-2.0 110M parameters 512 tokens transformers
View model