Tutorials for AI developers

The AI Developer Hub contains AMD ROCm tutorials in Jupyter Notebook format for training, fine-tuning, and inference. It leverages popular machine learning frameworks on AMD GPUs.

These tutorials are organized into four main categories:

• Inference: Resources for running inference with trained models.

• Fine-tuning: Examples and guides for fine-tuning machine learning models.

• Pretraining: Tutorials on pretraining models from scratch.

• GPU development and optimization: Resources for optimizing AI compute and kernel development on GPUs.

All tutorials on the AI Developer Hub are available to download in the Jupyter Notebook format from the public GitHub repository at ROCm/gpuaidev.

Inference tutorials

• AI agent with MCPs using vLLM and PydanticAI

• Hugging Face Transformers

• Hugging Face TGI

• Deploying with vLLM

• From chatbot to rap bot with vLLM

• RAG with LlamaIndex and Ollama

• OCR with vision-language models with vLLM

• Building AI pipelines for voice assistants

• Speculative decoding with vLLM

• Llama Stack

• DeepSeek-R1 with SGLang

Fine-tuning tutorials

• VLM with PEFT

• LLM with LoRA

• LLM with QLoRA

• Llama-3.1 8B with torchtune

• Llama-3.1 8B with Llama-Factory

• GRPO with Unsloth

Pretraining tutorials

• OLMo model with PyTorch FSDP

• Training configuration with Megatron-LM

• LLM with Megatron-LM

• Llama-3.1 8B with torchtitan

• Custom diffusion model with PyTorch

GPU development and optimization tutorials

• Kernel development and optimization with Triton

• Profiling Llama-4 inference with vLLM

• FP8 quantization with AMD Quark for vLLM

To contribute to the documentation, see Contributing to ROCm.

You can find ROCm licensing information on the Licensing page.