deepreinforce-ai/CUDA-L2
A system that uses reinforcement learning agents and LLMs to automatically tune Half-precision General Matrix Multiply (HGEMM) CUDA kernels for GPUs.
Not currently ranked — collecting fresh signals.
star history
CUDA-L2 combines large language models with reinforcement learning to search and optimize CUDA kernel configurations for matrix multiplication. The system automatically outperforms established baselines including NVIDIA’s cuBLAS library and cuBLASLt auto-tuning by generating optimized kernel parameters. It supports multiple GPU architectures including RTX 3090, A100, and H100, targeting the core computational primitive that underlies LLM and deep learning inference.
Frequently asked
- What is deepreinforce-ai/CUDA-L2?
- A system that uses reinforcement learning agents and LLMs to automatically tune Half-precision General Matrix Multiply (HGEMM) CUDA kernels for GPUs.
- Is CUDA-L2 open source?
- Yes — deepreinforce-ai/CUDA-L2 is open source, released under the MIT license.
- What language is CUDA-L2 written in?
- deepreinforce-ai/CUDA-L2 is primarily written in Cuda.
- How popular is CUDA-L2?
- deepreinforce-ai/CUDA-L2 has 459 stars on GitHub.
- Where can I find CUDA-L2?
- deepreinforce-ai/CUDA-L2 is on GitHub at https://github.com/deepreinforce-ai/CUDA-L2.