UI-MOPD/Qwen3-VL-8B-Thinking-Desktop-SFT

VISIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 4, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

UI-MOPD/Qwen3-VL-8B-Thinking-Desktop-SFT is an 8 billion parameter multimodal model, fine-tuned from Qwen3-VL-8B-Thinking, specifically for desktop GUI interaction tasks. Developed by UI-MOPD, it excels as a desktop GUI agent for executing tasks like file management and application control. This model serves as an auxiliary artifact to demonstrate the quality of Uni-GUI SFT data and facilitate model-merge experiments within the 8B GUI agent tier.

Loading preview...

Overview

UI-MOPD/Qwen3-VL-8B-Thinking-Desktop-SFT is an 8 billion parameter multimodal model, supervised fine-tuned (SFT) from the Qwen3-VL-8B-Thinking base model. It has been trained on desktop interaction trajectories derived from the Uni-GUI dataset, which focuses on OSWorld-style desktop environments. This model is released as an auxiliary artifact by the UI-MOPD project to showcase the effectiveness of the Uni-GUI SFT data at the 8B scale and to support community efforts in model merging (e.g., TIES/DARE) with other 8B GUI agents.

Key Capabilities & Features

  • Desktop GUI Agent: Designed to act as a standalone agent for interacting with desktop graphical user interfaces.
  • Multimodal: Inherits multimodal capabilities from its Qwen3-VL-8B-Thinking base, enabling it to process visual inputs from desktop environments.
  • SFT on Uni-GUI Data: Fine-tuned specifically on high-quality desktop interaction data, providing a strong baseline for desktop task execution.
  • Reference for Data Quality: Serves as a benchmark to evaluate the quality and impact of the Uni-GUI SFT data on 8B models.
  • Enables Model-Merge Experiments: Provides a suitable base for researchers to experiment with model merging techniques (TIES, DARE) with other 8B GUI agents.

Relationship to UI-MOPD Pipeline

It's important to note that this SFT model is not part of the primary UI-MOPD student training pipeline. In the main UI-MOPD approach, the 8B student model is cold-started directly from the Qwen3-VL-8B-Thinking base and trained via on-policy distillation from larger 32B teacher models. This Desktop-SFT model is an independent release for community use and research.

Intended Use Cases

  • Developing and testing desktop GUI automation solutions.
  • Researching model merging strategies for GUI agents.
  • Establishing a baseline for SFT-only performance compared to distillation-enhanced models.
  • Evaluating the impact of Uni-GUI dataset on 8B multimodal models.

Finetunes
1 models