GLM-5.3-Flash-oQ3e-mtp
[!NOTE] Updated 2026-10-24: Re-quantized with Better Low-Bit oQ. Please re-download if you have an earlier copy.
This model was quantized using oQ (oMLX v0.7.3.dev1) mixed-precision quantization with Better Low-Bit oQ (jundot/omlx#4776).
A mixed-precision MLX checkpoint of zai-org/GLM-5.3-Flash, a natively multimodal mixture-of-experts model with 320B total and 18B active parameters that combines sparse and linear attention with Manifold-Constrained Hyper-Connections (mHC). It includes the language model, the vision encoder and the multi-token prediction (MTP) head.
The effective average is 3.694 bits/weight for the complete checkpoint.
Quantization and Bit Distribution
oQ3e allocates bits by measured layer sensitivity, then rounds each group with an importance matrix (128 samples x 512 tokens) and a weighted least-squares refit of its scale and bias.
| Storage format | Logical weights | Share of total | Tensor storage | Effective bits/weight |
|---|---|---|---|---|
| Affine 3-bit, group 64 | 304.406B | 94.735% | 124.031 GiB | 3.50 |
| Affine 4-bit, group 64 | 7.248B | 2.256% | 3.797 GiB | 4.50 |
| Affine 8-bit, group 64 | 8.976B | 2.793% | 8.882 GiB | 8.50 |
| BF16 | 0.607B | 0.189% | 1.131 GiB | 16.00 |
| FP32 | 0.086B | 0.027% | 0.322 GiB | 32.00 |
| Total | 321.323B | 100% | 138.163 GiB | 3.694 |
Storage Breakdown
| Component | Logical weights | Storage |
|---|---|---|
| Language backbone | 313.327B | 142.915 GB / 133.100 GiB |
| Vision encoder | 0.564B | 1.127 GB / 1.050 GiB |
| MTP head | 7.433B | 4.309 GB / 4.013 GiB |
| Total | 321.323B | 148.351 GB / 138.163 GiB |
License
MIT, following the source model. See LICENSE.