Name the MLX-layout safetensors in the title and first paragraph. Document Metal as the default on macOS, the 16-step default and Fast/Balanced/Best presets, and add measured speed, memory and accuracy (PyTorch parity, Metal vs CPU, step study). manifest.json carries the new README hash. Weights are unchanged.
Expand the model card and add params to the manifest
Explain what TurboSpark is, why this conversion exists, that the files are MLX-layout safetensors rather than an MLX model, what each file does, how it was converted, and how to use it. manifest.json gains parameter counts and the new README hash. Weights are unchanged.
