Smart-Solidity-beta Overview

Smart Solidity beta is a fine-tuned version of CodeLlama-7B-Instruct, tailored for generating and understanding Solidity smart contract code. This model specializes in producing high-quality Solidity code based on user-provided instructions and use cases. It was fine-tuned to ensure robust performance and precise outputs in the blockchain and smart contract domain.

Key Features

Language: Trained exclusively for Solidity and related blockchain development.
Purpose: Tailored for creating, debugging, and understanding Solidity smart contracts.
Ease of Use: Provides concise and accurate responses to Solidity-specific queries.

Use Cases

Code Generation: Generate boilerplate or advanced Solidity code snippets for smart contracts.
Code Explanation: Understand complex Solidity logic by receiving step-by-step explanations.
Debugging: Identify and suggest fixes for potential bugs or inefficiencies in smart contract code.
Optimization: Propose refactored versions of Solidity code for gas efficiency and maintainability.
Learning: Assist blockchain developers in learning Solidity through practical examples.

Model Details

Base Model: Meta’s CodeLlama-7b.
Fine-tuned Data: Processed dataset containing GPT-generated human instruction and Solidity source code data pairs.
Model Size: 7 billion parameters, balancing high-quality output with reasonable computational requirements.

Technical Specifications

Input Format: Accepts Solidity code snippets, prompts, or questions in natural language.
Output Format: Provides Solidity code, recommendations, or explanations in plain text.

Training Loss Table

Step	Training Loss
100	0.3309
1000	0.2995
2000	0.2275
3000	0.2695
4000	0.2514
5000	0.2405

Hardware:

GPU: 1x NVIDIA GeForce GTX 1080Ti
Training time: ~48 hours

Example Usage

The following Python code demonstrates how to use the EclipseNomad/Smart-Solidity-beta model to generate Solidity smart contract code:

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Load the tokenizer
tokenizer = AutoTokenizer.from_pretrained("EclipseNomad/Smart-Solidity-beta")

# Load the model with 4-bit quantization
model = AutoModelForCausalLM.from_pretrained(
    "EclipseNomad/Smart-Solidity-beta",
    load_in_4bit=True,
    device_map="auto"
)

# Define the instruction for the model
instruction = (
    "Create a smart contract to manage a whitelist of wallet addresses. "
    "Include functionality for a DAO to approve/revoke wallets and a mechanism to set a validator address."
)

# Tokenize the instruction
input_ids = tokenizer(instruction, return_tensors="pt").input_ids.to("cuda")

# Generate Solidity code
outputs = model.generate(
    input_ids=input_ids,
    max_new_tokens=512,
    temperature=0.7,
    top_p=0.9,
    pad_token_id=tokenizer.pad_token_id,
)

# Decode and print the generated Solidity code
solidity_code = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(solidity_code)



# Limitations
The model may occasionally produce incorrect or suboptimal Solidity code; thorough human review is recommended before deploying to production.
Its training data might not encompass the most recent Solidity updates or EVM standards. Ensure compatibility with the latest versions.
Future Work
The model is open to further fine-tuning and community contributions to enhance its accuracy and support for emerging Solidity standards and advanced blockchain use cases.