Port PyTorch / CUDA diffusion, transformer, LLM, audio, or 3D models to Apple MLX for Apple-Silicon inference. Invoke whenever the user is working on an MLX port — scaffolding a `-mlx` fork, translating attention / RoPE / VAE / norm layers, setting up PyTorch vs MLX parity tests, diagnosing wrong numerics (black images, cyan textures, gray output, garbage tokens, shape-safe silent failures), picki
Desenvolvimento#llm#aipor dgrauet