DolphinCoder StarCoder2 15B
| Property | Value |
|---|---|
| License | bigcode-openrail-m |
| Training Infrastructure | 8x H100s with qLoRA and Axolotl |
| Training Duration | 3 days (3 epochs) |
| Primary Language | English |
What is dolphincoder-starcoder2-15b?
DolphinCoder is a specialized AI programming assistant built on the StarCoder2-15b architecture, specifically optimized for software engineering tasks. Trained on a diverse set of 8 high-quality datasets including Magicoder, OpenHermes, and Code-Feedback, this model excels at generating, analyzing, and modifying code across various programming languages.
Implementation Details
The model utilizes the ChatML prompt format for interactions and was trained using qLoRA and Axolotl frameworks. It's available in multiple quantized versions including GGUF and ExLlamaV2 formats for efficient deployment.
- Comprehensive training on 8 specialized coding datasets
- Uncensored architecture with filtered dataset to remove alignment bias
- Supports multiple programming languages and software engineering tasks
- Implements ChatML prompt format for structured interactions
Core Capabilities
- Advanced code generation and problem-solving
- Multi-language programming support
- Code analysis and optimization
- Highly compliant response generation
- Efficient processing through quantized versions
Frequently Asked Questions
Q: What makes this model unique?
The model's unique strength lies in its specialized training on multiple coding datasets and its uncensored nature, making it highly adaptable to various programming tasks while maintaining high compliance with user requests.
Q: What are the recommended use cases?
The model is ideal for software development tasks, code generation, debugging, and programming education. However, users should implement their own alignment layer before deploying it as a service due to its uncensored nature.
