Save on courses or Exam now
About Instructors
-
edwardLary- Instructor Rating - Reviews - Students - CoursesNo biography available, please complete
🚀 Course Goals
- Install and operate Qwen locally on a multi-GPU server, and pick the right variant and quantization
- Write prompts that hold up: roles, system messages, few-shot, chaining and validated JSON output
- Analyse PDFs, scans, charts, invoices and video frames, and produce reports from them
📘 About This Course
This course is about running generative AI on hardware you control, and then making it do real work. Qwen runs locally on your own GPUs throughout: nothing is sent to a hosted API, and every lesson is a demonstration on a real multi-GPU server rather than a slide about one.
You start by installing the model family and choosing between variants, then learn to write prompts that hold up in production: structured roles, system messages, few-shot examples, task chaining and JSON output you can actually validate. From there you handle real documents - PDFs, scans, charts, invoices, video frames - and turn them into reports. Then you wire Qwen into an open-source automation platform, build multi-tool agents with memory and human approval, and assemble a complete multimodal RAG with chunking, hybrid search, reranking and citations you can trace back to a page.
The last three modules make it durable: QLoRA fine-tuning on a single GPU, an evaluation and security layer with prompt-injection testing and PII masking, and finally containerised deployment serving several users across four GPUs, with quotas, backups and monitoring. The course ends with a final project you build and present.
What You'll Learn
Install and operate Qwen locally on a multi-GPU server, and pick the right variant and quantization
Write prompts that hold up: roles, system messages, few-shot, chaining and validated JSON output
Analyse PDFs, scans, charts, invoices and video frames, and produce reports from them
Automate real business processes with an open-source platform wired to a local Qwen API
Build multi-tool AI agents with state, memory, permissions and human approval
Build a complete multimodal RAG: chunking, embeddings, hybrid search, reranking and citations
Fine-tune with LoRA and QLoRA on a single GPU, then evaluate, save and serve the adapter
Evaluate, secure and govern the system, then containerise and deploy it for several users
🛠️ Requirements
- A machine with at least one NVIDIA GPU; the course server uses four
- Comfort with the Linux command line, basic Python and Docker
- No prior experience with large language models is required
- Everything is open source: no paid API key and no cloud account needed
- Disk space for model weights and a local vector database
👤 Who This Course is For
- Developers and IT professionals who want to run generative AI on their own infrastructure
- Automation and data engineers building document and process pipelines
- Teams that cannot send internal data to a third-party API
- Anyone who would rather run and measure a model than call a hosted endpoint
53618 Students applied
100 Comments
-
-
John DoeMorbi nec nisi ante. Quisque lacus ligula, iaculis in elit et, interdum semper quam. Fusce in interdum tortor. Ut sollicitudin lectus dolor eget imperdiet libero pulvinar sit amet
-
John Doe
Morbi nec nisi ante. Quisque lacus ligula, iaculis in elit et, interdum semper quam. Fusce in interdum tortor. Ut sollicitudin lectus dolor eget imperdiet libero pulvinar sit amet
-
-
John Doe
Morbi nec nisi ante. Quisque lacus ligula, iaculis in elit et, interdum semper quam. Fusce in interdum tortor. Ut sollicitudin lectus dolor eget imperdiet libero pulvinar sit amet
-
John Doe
Morbi nec nisi ante. Quisque lacus ligula, iaculis in elit et, interdum semper quam. Fusce in interdum tortor. Ut sollicitudin lectus dolor eget imperdiet libero pulvinar sit amet
-
John Doe
Morbi nec nisi ante. Quisque lacus ligula, iaculis in elit et, interdum semper quam. Fusce in interdum tortor. Ut sollicitudin lectus dolor eget imperdiet libero pulvinar sit amet
Morbi nec nisi ante. Quisque lacus ligula, iaculis in elit et, interdum semper quam. Fusce in interdum tortor. Ut sollicitudin lectus dolor eget imperdiet libero pulvinar sit amet