DDIDU/ETRI_CodeLLaMA_7B_CPP

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Mar 14, 2024License:llama2Architecture:Transformer0.0K Open Weights Cold

DDIDU/ETRI_CodeLLaMA_7B_CPP is a 7 billion parameter CodeLLaMA-based model developed by DDIDU and ETRI, specifically pre-trained and fine-tuned for C++ code generation. It leverages LoRa for further pre-training on 543 GB of C++ code and fine-tuning on CodeM's C++ instruction data. This model excels at C++ programming tasks, achieving a pass@1 of 34.3% on HumanEvalX-cpp, making it highly specialized for C++ code completion and generation.

Loading preview...

ETRI_CodeLLaMA_7B_CPP: C++ Specialized Code Generation

This model, developed by DDIDU and ETRI, is a 7 billion parameter variant of Meta's CodeLLaMA, uniquely specialized for C++ programming. It underwent extensive further pre-training using LoRa on a massive 543 GB dataset of high-quality C++ code collected online. Following this, it was fine-tuned with CodeM's C++ instruction data to enhance its ability to understand and generate C++ code based on instructions.

Key Capabilities

  • C++ Code Generation: Highly optimized for generating C++ code, leveraging its specialized training.
  • Instruction Following: Fine-tuned on C++ instruction data, enabling it to respond effectively to C++ programming prompts.
  • Performance: Achieved a pass@1 score of 34.3% on the HumanEvalX-cpp benchmark, demonstrating its proficiency in C++ coding tasks.

Good For

  • Developers requiring a dedicated model for C++ code completion and generation.
  • Applications focused on C++ programming assistance and automated code writing.
  • Research and development in C++-specific large language models.