amkyawdev/amk-coder-v2
amkyawdev/amk-coder-v2 is a 2 billion parameter Qwen2.5-Coder-1.5B based causal language model, fine-tuned by amkyawdev using LoRA (PEFT) for coding assistance. It features a 32,768 token context length and specializes in code generation, debugging, and explanation with full support for both Burmese and English. This model is optimized for developers requiring a localized coding agent for Myanmar-specific contexts.
Loading preview...
amk-coder-v2: Myanmar Coding Agent
amk-coder-v2 is a 2 billion parameter coding assistant developed by amkyawdev, fine-tuned from Qwen2.5-Coder-1.5B using LoRA (PEFT). This model is specifically localized for Myanmar, offering comprehensive support for Myanmar Unicode text alongside English. It boasts a substantial context length of 32,768 tokens, making it suitable for handling complex coding tasks.
Key Capabilities
- Myanmar Language Support: Full integration for Myanmar Unicode in coding contexts.
- Multi-language Code Generation: Generates code in Python, JavaScript, C++, Java, and other languages.
- Debugging: Identifies and suggests fixes for bugs in code.
- Code Explanation: Provides line-by-line explanations of code snippets.
- Efficient Training: Fine-tuned on the
amkyawdev/mm-llm-coder-agent-dataset(4M rows) over 3 epochs using paged_adamw_8bit optimizer on Kaggle Dual T4 GPUs.
Good For
- Developers in Myanmar needing a coding assistant that understands and generates code in both Burmese and English.
- Automating code generation and debugging for common programming languages.
- Educational purposes, providing explanations for code logic.
Limitations
Users should be aware of the 32,768 token context limit, the necessity to verify generated code for quality, and potential issues with Myanmar Unicode if not properly converted. The model's domain knowledge is primarily limited to common programming languages.