Dolphy-1.0: A Compact Yet Powerful Language Model
Dolphy-AI/Dolphy-1.0 represents Dolphy AI's initial foray into machine learning, building upon the robust Qwen3 4B 2507 Instruct base model. This 4 billion parameter model has undergone extensive fine-tuning using Unsloth LoRA, leveraging 1.5 million carefully curated examples across 20 diverse datasets. The meticulous training process aims to enhance Qwen's inherent capabilities, positioning Dolphy-1.0 as a strong performer within the 4B parameter class.
Key Capabilities
- Enhanced Performance: Fine-tuned on a large and diverse dataset to outperform many larger models in its category.
- Qwen3 Compatibility: Inherits full compatibility with Qwen3's advanced features, including extensive tool use, function calling, and robust multilingual support.
- Unchanged Architecture: Maintains the original Qwen3 tokenizer and model architecture, ensuring seamless integration and consistent behavior.
- Ready for Inference: Designed for immediate use without additional pre-inference steps, with GGUF models recommended for local, fast inference.
Ideal Use Cases
Dolphy-1.0 is particularly well-suited for applications requiring a lightweight yet capable language model. Its strong performance in the 4B category, combined with Qwen3's inherent features, makes it an excellent choice for:
- General-purpose conversational AI.
- Applications requiring function calling and tool use.
- Multilingual text generation and understanding tasks.
- Scenarios where resource efficiency is critical without compromising on capability.