Skip to content

STetsing/solcoder

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SolCoder

SolCoder is LLM which generates solidity source code from user prompt comments. It is hosted on the Huggingface repository Pipper/Solcoder and is continuously uptated. The model output is focussed on generating functions from user comments.

P.S.: No source code context is considered if using an editor or such. The generated code might not be executable due to context variables and other dependencies, that might be defined prior to the generation. The user might adjust these to be aligned with the context.

Dataset Sources for Training & Finetuning

The datasets used is an aggregation of multiple datasources.

Sourcify

You can follow the instructions in the docs and contact Kaan Uzdogan for the credentials.

See slither-solidity for the sourcify data processing.

Slither audited smart contracts

mwritescode/slither-audited-smart-contracts

StarCoder

bigcode/the-stack-dedup

from datasets import load_dataset

# specific language (e.g. Dockerfiles) 
ds = load_dataset("bigcode/the-stack-dedup", data_dir="data/solidity", split="train")

Audit and Security

IMPORTANT: None of the code parsed in the dataset is statically audited or further audited if already. Use static analyzers such as slither for the purpose.

Setup the Hugginface Tokens

Important: The token must have write access. This is use to push data and model to the HF hub automatically at creation time

export HF_TOKEN="your write access token"
export HF_BEARER_TK="your api bearer token from HF"

Launch accelerated training

accelerate launch --num_cpu_threads_per_process 8 train/train_phi2_lora_qa.py

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published