ByteDance-Seed/cudaLLM-8B

cudaLLM-8B by ByteDance-Seed is an 8 billion parameter language model based on Qwen3-8B, specifically designed for generating high-performance and syntactically correct CUDA kernels. It underwent a two-stage training process to master parallel programming for GPUs, achieving notable performance on the KernelBench dataset. This model excels at assisting developers in writing and optimizing CUDA code for scientific computing and machine learning workloads.