tag
#finetuning
2 posts
CUDA Agent paper notes
Notes on CUDA Agent and how it uses an agentic RL workflow with tools for CUDA kernel optimization.
ReFT: The Future of LLM Fine-Tuning
A new method called Reinforced Fine-Tuning (ReFT) uses reinforcement learning to help large language models solve complex math problems more effectively than traditional supervised fine-tuning.