Model: JetBrains/Mellum2.1-12B-A2.5B-Thinking
Source revision: 92ddae9fc7665e9f801d141d2e5a6b2caf2460c4
Original model created and trained by JetBrains; Apache-2.0.
Modified artifact: native MLX affine 6-bit/g64 quantization with upstream 8-bit router policy.
Conversion and independent validation: Imad, https://github.com/imaddde867/mellum-mlx
No fine-tuning or calibration applied.