Model Overview
nightmedia/Qwen3-4B-Agent-Eva is a 4 billion parameter model, created by merging Qwen3-4B-Agent and FutureMa/Eva-4B. This merge aims to combine the agentic capabilities of Qwen3-4B-Agent with the conversational skills introduced by Eva-4B, which was originally designed for detecting evasive answers. The model maintains a high performance level, with its qx86-hi quantization performing comparably to the full precision version.
Key Capabilities
- Agentic Behavior: The model is specifically profiled to act as agents, particularly in roleplay scenarios such as Star Trek DS9 station characters.
- Conversational Skills: Incorporates conversational abilities from the Eva-4B component, enhancing its interactive dialogue.
- General Tasks: While specialized for agentic roleplay, it can also be utilized for regular language model tasks.
- Performance: Achieves performance metrics (e.g., 0.568, 0.775, 0.872, 0.699, 0.418, 0.777, 0.654) that are noted to be competitive with larger models.
Good For
- Roleplaying: Ideal for creating interactive agents in specific narrative settings, like the Star Trek DS9 universe.
- Conversational AI: Suitable for applications requiring robust dialogue and interaction.
- Agent-based Simulations: Can be used in scenarios where models need to exhibit defined agentic characteristics and accountability.