Daniel Han on The Future of Training and Reinforcement Learning

Join Daniel Han, CEO of Unsloth, as he shares insights on optimizing AI models onstage at AMD Advancing AI 2025. Unsloth, is an open-source AI startup on a mission to make AI accessible and accurate for everyone. In this session, Daniel discusses the latest advancements in training and reinforcement learning, including fixing gradient accumulation bugs and introducing async offloaded gradient checkpointing.

Learn how Unsloth collaborates with industry giants like Hugging Face, Google, and Meta to enhance open-source models including Llama and Mistral. Get practical tips on running large models on local devices using AMD GPUs and the benefits of dynamic quantization. This talk is a must-watch for AI developers interested in model performance and efficiency.

Train your own R1 reasoning model with Unsloth: https://rocm.docs.amd.com/projects/ai-developer-hub/en/latest/notebooks/fine_tune/unsloth_Llama3_1_8B_GRPO.html

Find the resources you need to develop using AMD products: https://www.amd.com/en/developer.html

***

© 2025 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, EPYC, ROCm, and AMD Instinct and combinations thereof are trademarks of Advanced Micro Devices, Inc.

Видео Daniel Han on The Future of Training and Reinforcement Learning канала AMD Developer Central