yeonjooooni/Gemma3-4B_WEASEL
yeonjooooni/Gemma3-4B_WEASEL is a 4.3 billion parameter language model, fine-tuned from Google's Gemma-3-4b-it. This WEASEL variant is specifically optimized for web-agent style reasoning and action generation. It excels at web task solving through instruction following and generating reasoning traces, making it suitable for research in web-agent behavior.
Loading preview...
Overview
The yeonjooooni/Gemma3-4B_WEASEL model is a specialized 4.3 billion parameter language model, built upon the google/gemma-3-4b-it base. It has been fine-tuned using the WEASEL method to enhance its capabilities in web-agent style reasoning and action generation. The model's primary objective is to facilitate web task solving by generating appropriate reasoning and actions based on supervision.
Key Capabilities
- Web-Agent Reasoning: Designed to generate logical reasoning traces for web-based tasks.
- Action Generation: Capable of suggesting actions relevant to web environments.
- Instruction Following: Optimized to interpret and execute instructions within a web-agent context.
Training and Intended Use
The model was fine-tuned on an AgentTrek-style web-agent dataset, focusing on web task solving with reasoning/action style supervision. It is intended for research purposes, particularly in exploring web-agent behavior, instruction following, and the generation of reasoning traces. Developers can load it using Hugging Face Transformers with AutoProcessor and AutoModelForImageTextToText.
Limitations
Users should be aware that the model may generate incorrect or unsafe browser actions. Its reasoning traces do not guarantee factual correctness, and it has not been evaluated for safety-critical applications.