Overview

The beachcities/qwen3-4b-sft-dpo-v25mix-structeval is a 4 billion parameter language model developed by beachcities. It is a fine-tuned variant of the unsloth/Qwen3-4B-Instruct-2507 base model, leveraging the Qwen3 architecture. A key characteristic of this model is its training methodology, which utilized Unsloth and Huggingface's TRL library, resulting in a reported 2x acceleration in training speed. This model is designed for tasks benefiting from its 40960 token context window.

Key Capabilities

• Efficient Training: Benefits from Unsloth's optimizations for faster fine-tuning.
• Extended Context: Supports a 40960 token context length, enabling processing of longer inputs and generating more coherent, extended outputs.
• Qwen3 Architecture: Built upon the robust Qwen3 foundation, providing strong general language understanding and generation abilities.

Good for

• Applications requiring a compact yet capable language model with a very large context window.
• Scenarios where efficient fine-tuning is a priority.
• Tasks involving long-form text analysis, summarization, or generation.