razxr/qwen3-vl-4b-2294-project_v4

VISIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:May 18, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

The razxr/qwen3-vl-4b-2294-project_v4 is a 4 billion parameter Qwen3-VL model, developed by razxr, fine-tuned for specific tasks. This model was optimized for faster training using Unsloth and Huggingface's TRL library. It is designed for applications requiring efficient performance from a vision-language model.

Loading preview...

Overview

This model, razxr/qwen3-vl-4b-2294-project_v4, is a 4 billion parameter Qwen3-VL instruction-tuned model developed by razxr. It was fine-tuned from the unsloth/Qwen3-VL-4B-Instruct base model.

Key Characteristics

  • Architecture: Based on the Qwen3-VL family, indicating vision-language capabilities.
  • Parameter Count: Features 4 billion parameters, offering a balance between performance and computational efficiency.
  • Training Optimization: The model was fine-tuned using Unsloth and Huggingface's TRL library, which enabled a 2x faster training process.
  • License: Distributed under the Apache-2.0 license.

Use Cases

This model is suitable for applications that benefit from a vision-language model, particularly where efficient training and deployment are critical. Its optimized training process suggests it can be a good choice for developers looking for a performant Qwen3-VL variant with reduced training overhead.

Finetunes
1 models