langfeng01/GiGPO-Qwen2.5-7B-Instruct-WebShop
langfeng01/GiGPO-Qwen2.5-7B-Instruct-WebShop is a 7.6 billion parameter instruction-tuned model based on the Qwen2.5 architecture, specifically trained using the GiGPO method. This model is specialized for autonomous agent operation within the WebShop e-commerce environment, designed to reason and select actions to achieve shopping goals. It features a 131072 token context length, making it suitable for complex, multi-step interactive tasks in web-based environments.
Loading preview...
Overview
GiGPO-Qwen2.5-7B-Instruct-WebShop is a 7.6 billion parameter language model built upon the Qwen2.5-7B-Instruct architecture. It has been specifically fine-tuned using the GiGPO (Generative Imitation from Guided Policy Optimization) method, as detailed in the associated arXiv paper. This model is designed for autonomous agent tasks within the WebShop e-commerce environment, enabling it to navigate, reason, and interact to fulfill shopping objectives.
Key Capabilities
- Autonomous Agent Operation: Specialized in acting as an expert agent within the WebShop environment.
- Step-by-Step Reasoning: Employs a
<think>tag for detailed internal reasoning before taking an action. - Action Selection: Selects appropriate actions from a given set of admissible options, enclosed within
<action>tags. - Contextual Awareness: Utilizes prompt templates that incorporate task descriptions, current observations, available actions, and historical interactions to inform decision-making.
Good For
- Web-based Agent Development: Ideal for researchers and developers working on agents that interact with e-commerce platforms or similar web interfaces.
- Reinforcement Learning from Human Feedback (RLHF) Research: Demonstrates an application of the GiGPO training methodology for complex interactive tasks.
- Simulated E-commerce Tasks: Suitable for automating shopping processes, product search, and other related activities in a simulated environment.