claudegpt code logic debugger v0.1

Introduction

ClaudeGPT Code Logic Debugger v0.1 is designed for code generation, debugging, and editing. It aims to achieve performance on par with or superior to GPT-4o and Claude-3.5 in specific tasks, providing an alternative to cloud-based services to address rate limits and privacy concerns.

Architecture

The model is optimized for inference speed on hardware with at least 24 GB VRAM, such as RTX 3090. It supports complex debugging scenarios, particularly where multi-library dependencies are involved, and offers a structured approach for identifying and resolving issues.

Training

The model's performance has been evaluated in programming tasks like debugging and generation. It features different modes—Balanced and Precise—to cater to various use cases, adjusting parameters like max_tokens, temperature, top_k, and top_p to optimize performance.

Guide: Running Locally

1. Prerequisites: Ensure your system has a GPU with at least 24 GB VRAM.
2. Installation: Install the Hugging Face CLI tool:

   pip install -U "huggingface_hub[cli]"
   

3. Download the Model:
   • For commercial use:

     huggingface-cli download FredZhang7/claudegpt-code-logic-debugger-v0.1 --include "AutoCoder.IQ4_K.gguf" --local-dir ./
     

   • For non-commercial use:

     huggingface-cli download FredZhang7/claudegpt-code-logic-debugger-v0.1 --include "codestral-22b-v0.1-IQ6_K.gguf" --local-dir ./
     

4. Cloud GPUs: Consider using cloud services like AWS, Google Cloud, or Azure for access to powerful GPUs if local hardware is insufficient.

License

The model is available under the Apache 2.0 or MNPL 0.1 license. Codestral-22b-v0.1-IQ6_K.gguf must be used only for non-commercial projects. For commercial activities, use alternatives such as Qwen2-7b-Instruct bf16 or AutoCoder.IQ4_K.gguf.