Modern VLMs: Qwen, Gemma & Kimi Training in Australia

This program introduces participants to Vision-Language Models (VLMs) using fully open-source, modern models like Qwen 3 VL, Gemma 3 Vision, and Kimi-VL.
Learners will understand how multimodal AI works, prepare simple image/video datasets, run open-source VLMs, perform fine-tuning using LoRA, and deploy multimodal applications.

The course focuses on intuitive explanations, hands-on guided exercises, and easy-to-use tools (Google Colab, Hugging Face Spaces, Gradio).

By the end, learners will be able to build and deploy their own simple VLM applications.

Welcome to the official Modern VLMs: Qwen, Gemma & Kimi Training certification program. This comprehensive training is designed to elevate your professional skills and provide you with practical, industry-relevant knowledge in in Australia. As a globally recognized corporate training provider operating in 55+ countries, ProgNXT ensures that our curriculum meets the highest standards of excellence.

Whether you are looking to upskill your team or advance your personal career, our expert-led sessions will guide you through the core concepts of this domain. Upon successful completion of the 21 Hrs program, participants will receive a globally accepted certification, demonstrating their proficiency and readiness to tackle complex challenges in the field.

• Basic Python syntax (functions, lists, loops)
• Very basic understanding of machine learning concepts (optional but helpful)
• Ability to use Google Colab notebooks

Skillset Achieved After the Course

Vision-Language Understanding

• Understanding how AI interprets images, text, and video together
• Working with modern open-source VLMs

Data Handling

• Preparing image–text datasets
• Preparing small video–text datasets
• Basic image and frame preprocessing

Model Usage & Fine-Tuning

• Running Qwen 3 VL, Gemma 3, and Kimi-VL
• Performing simple LoRA fine-tuning on images
• Introductory video fine-tuning workflows

Evaluation & Debugging

• Identifying good vs bad model outputs
• Recognizing hallucinations and common model errors

Deployment

• Building a simple multimodal web app using Gradio
• Deploying on Hugging Face Spaces
• Running models on Google Colab (free)

Responsible AI Awareness

• Understanding bias, fairness, and privacy concerns in multimodal AI

Course Outcomes

By the end of the training, participants will be able to:

1. Understand key VLM concepts

• Explain what VLMs are
• Describe how vision encoders and language models work together
• Explain differences between Qwen 3 VL, Gemma 3, and Kimi-VL

2. Work with multimodal data

• Prepare image/text datasets
• Extract video frames and create small video caption datasets

3. Use and fine-tune open-source VLMs

• Load and experiment with VLMs on Colab
• Perform beginner-friendly LoRA-based fine-tuning
• Run inference on images and short videos

4. Deploy multimodal applications

• Build a simple UI that accepts images/videos and answers questions
• Deploy it publicly on Hugging Face Spaces

5. Understand responsible AI considerations

• Know where VLMs might hallucinate
• Recognize privacy and fairness concerns in vision/video AI