vanch007/Ming-3.0-tts-tiny-mlx-bf16
Pure Apple Silicon MLX native port of inclusionAI/Ming-3.0-tts-tiny (7.9B MoE with 128 Experts + 1 Shared Expert + KDA Attention) in bfloat16 Full Precision.
- Architecture: 128 Routed Experts + 1 Shared Expert + KDA Recurrent Linear Attention
- Size: 17.96 GB
- Zero Triton: Pure Metal acceleration on Apple Silicon (M1/M2/M3/M4)