sabi-noema-oQ4e-mtp
This model was quantized from sabicap/sabi-noema using oQ (oMLX v0.7.1rc1) mixed-precision quantization.
The MTP head was grafted from Jundot/Qwen3.8-27B-oQ4e-mtp.
The effective average is 4.889 bits/weight for the complete checkpoint.
Quantization and Bit Distribution
| Storage format | Logical weights | Share of total | Tensor storage | Effective bits/weight |
|---|---|---|---|---|
| Affine 4-bit, group 64 | 22.690B | 81.674% | 11.887 GiB | 4.50 |
| Affine 5-bit, group 64 | 4.287B | 15.431% | 2.745 GiB | 5.50 |
| Affine 6-bit, group 64 | 0.288B | 1.038% | 0.218 GiB | 6.50 |
| BF16 | 0.516B | 1.857% | 0.961 GiB | 16.00 |
| Total | 27.781B | 100% | 15.811 GiB | 4.889 |
Storage Breakdown
| Component | Logical weights | Storage |
|---|---|---|
| Language model | 26.896B | 15.741 GB / 14.660 GiB |
| Vision | 0.461B | 0.921 GB / 0.858 GiB |
| MTP head | 0.425B | 0.314 GB / 0.293 GiB |
| Total | 27.781B | 16.977 GB / 15.811 GiB |