Specializing a pre-trained Model

HuggingFace and similar hubs for models with open trained weights are a great way to get started, however, it is likely that these models will not perform as well for your particular data.

In this section, we will focus on specializing pre-trained models with your custom data. We will target AMD CPU and GPU.

Technologies covered: CPU, GPU.

Requisites

GPUs: AMD Instinct™ Accelerators, AMD Radeon™ RX Graphics Cards or AMD Radeon™ PRO Graphics Cards

Setup environment

Examples

• Re-training a model using PyTorch and Transfer Learning
• DistilBERT for Sentiment Analysis
• Fine-tune Llama 3.2 to generate Markdown friendly Python functions

Resources

• Retrieval Augmented Generation (RAG) using LlamaIndex

• How to enable RAG (Retrieval Augmented Generation) on an AMD Ryzen™ AI PC or Radeon Graphics Card

• How to get an AI coding assistant on your AMD Ryzen™ AI PC or Radeon Graphics Card

If you would like to suggest a resource, email us aup@amd.com

---

Copyright (C) 2025 Advanced Micro Devices, Inc. All rights reserved.

SPDX-License-Identifier: MIT