metadata

license: mit
language:
  - en
inference: true
base_model:
  - microsoft/codebert-base-mlm
  - web3se/SmartBERT-v2
pipeline_tag: fill-mask
tags:
  - fill-mask
  - smart-contract
  - web3
  - software-engineering
  - embedding
  - codebert

SmartBERT V3 CodeBERT

Overview

SmartBERT V3 is a pre-trained programming language model, initialized with CodeBERT-base-mlm. It has been further trained on SmartBERT V2 with an additional 64,000 smart contracts, to enhance its robustness in representing smart contract code at the function level.

Training Data: Trained on a total of 80,000 smart contracts, including 16,000 from SmartBERT V2 and 64,000 (starts from 30001) new contracts.
Hardware: Utilized 2 Nvidia A100 80G GPUs.
Training Duration: Over 30 hours.
Evaluation Data: Evaluated on 1,500 (starts from 96425) smart contracts.

Usage

from transformers import RobertaTokenizer, RobertaForMaskedLM, pipeline

model = RobertaForMaskedLM.from_pretrained('web3se/SmartBERT-v3')
tokenizer = RobertaTokenizer.from_pretrained('web3se/SmartBERT-v3')

code_example = "function totalSupply() external view <mask> (uint256);"
fill_mask = pipeline('fill-mask', model=model, tokenizer=tokenizer)

outputs = fill_mask(code_example)
print(outputs)

Preprocessing

All newline (\n) and tab (\t) characters in the function code were replaced with a single space to ensure consistency in the input data format.

Base Model

Original Model: CodeBERT-base-mlm

Training Setup

training_args = TrainingArguments(
  output_dir=OUTPUT_DIR,
  overwrite_output_dir=True,
  num_train_epochs=20,
  per_device_train_batch_size=64,
  save_steps=10000,
  save_total_limit=2,
  evaluation_strategy="steps",
  eval_steps=10000,
  resume_from_checkpoint=checkpoint
)

How to Use

To train and deploy the SmartBERT V3 model for Web API services, please refer to our GitHub repository: web3se-lab/SmartBERT.

web3se
/

SmartBERT-v3