zqmalyssa/Qwen2.5-1.5B-Assistant
zqmalyssa/Qwen2.5-1.5B-Assistant is a 1.5 billion parameter language model, fine-tuned from the Qwen2.5-1.5B architecture. This model has undergone Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) alignment, making it suitable for assistant-like conversational tasks. With a context length of 32768 tokens, it is optimized for generating coherent and contextually relevant responses in interactive applications.
Loading preview...
Model Overview
zqmalyssa/Qwen2.5-1.5B-Assistant is a compact yet capable language model, built upon the Qwen2.5-1.5B base architecture. It features 1.5 billion parameters and supports a substantial context length of 32768 tokens, allowing it to process and generate longer, more complex interactions.
Key Capabilities
This model has been enhanced through a two-stage alignment process:
- Supervised Fine-Tuning (SFT): Initial fine-tuning achieved a final loss of 1.65, establishing a strong foundation for instruction following.
- Direct Preference Optimization (DPO): Further alignment using DPO resulted in 70.4% accuracies and margins of 1.022, indicating its ability to generate responses that align with human preferences.
Good For
Given its fine-tuning for assistant-like behavior, this model is well-suited for:
- Conversational AI: Engaging in dialogue and providing helpful responses.
- Instruction Following: Executing specific commands or answering questions based on given prompts.
- Lightweight Applications: Its 1.5B parameter count makes it efficient for deployment in scenarios where computational resources are a consideration, while still offering a robust context window.