Forge MCP Server

GPU kernel optimization - 32 swarm agents turn PyTorch into fast CUDA/Triton kernels on real datacenter GPUs with up to 14x speedup