CodeWeave-LlamaCode

A fine-tuned code generation model built on Meta's CodeLlama foundation, specialized for enterprise code development.

Model Description

CodeWeave-LlamaCode extends codellama/CodeLlama-7b-Instruct-hf with domain-specific fine-tuning for improved code quality and developer productivity.

Base Model

• Foundation: CodeLlama-7b-Instruct-hf from Meta AI
• Architecture: Llama 2-based transformer
• Parameters: 7B

Training Data

Fine-tuned using CodeWeave-Enterprise dataset containing:

• 40K enterprise code samples
• API integration patterns
• Security-focused code examples
• Documentation generation tasks

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("toolevalxm/CodeWeave-LlamaCode")
tokenizer = AutoTokenizer.from_pretrained("toolevalxm/CodeWeave-LlamaCode")

Evaluation Results

Benchmark	Score
HumanEval	70.1%
MBPP	65.8%

Acknowledgements

We thank Meta AI for developing the CodeLlama series.

License

The license for this model is llama2.