Solidity LLM

Solidity LLM

An open-source 2-billion-parameter language model fine-tuned specifically for Solidity smart contract generation. Hosted on HuggingFace, MIT-licensed, and completely free to run on your own hardware.

Key Facts

Model	Chain-GPT/Solidity-LLM
Parameters	2B
License	MIT
Cost	Free (self-hosted)
Platform	HuggingFace
Compilation Success Rate	83%

When to Use This vs. the Smart Contract Generator

	Solidity LLM	Smart Contract Generator
Cost	Free	1 credit per request
Hosting	Self-hosted (you need a GPU)	Cloud API
Model size	2B parameters	Larger, more capable model
Quality	Good (83% compilation rate)	Higher quality, production-tuned
Offline	Yes	No
Best for	Experimentation, CI pipelines, air-gapped environments	Production contracts, complex logic

Use the Solidity LLM when you need offline generation, want zero API costs, or are integrating into CI/CD pipelines. Use the Smart Contract Generator when you need higher-quality output for production contracts.

Quick Start -- Python

Install Dependencies

pip install transformers torch

Generate a Smart Contract

from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = "Chain-GPT/Solidity-LLM"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

prompt = "Create a Solidity smart contract for an ERC-20 token with a 2% burn mechanism"

inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(
    **inputs,
    max_new_tokens=1024,
    temperature=0.7,
    do_sample=True,
)

generated_code = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(generated_code)

With GPU Acceleration

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = "Chain-GPT/Solidity-LLM"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float16,
    device_map="auto",
)

prompt = "Write a Solidity contract for a multi-signature wallet requiring 3 of 5 approvals"

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(
    **inputs,
    max_new_tokens=1024,
    temperature=0.7,
    do_sample=True,
)

generated_code = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(generated_code)

Tips

• The 83% compilation success rate means roughly 1 in 5 outputs may need manual correction. Always compile and test the output.

• Pair the generated output with the Smart Contract Auditor to catch vulnerabilities before deployment.

• Use float16 and device_map="auto" for efficient GPU memory usage.

• The model works well for standard patterns (ERC-20, ERC-721, staking, vesting) but may struggle with highly novel or complex contract logic.

• For production-critical contracts, consider using the cloud-based Smart Contract Generator instead.