---

base_model: zai-org/GLM-5.3-BF16

base_model_relation: quantized

library_name: gguf

license: other

license_name: glm-5.3

pipeline_tag: text-generation

language:

- en

- zh

tags:

- gguf

- llama.cpp

- glm_moe_dsa

- imatrix

---

GLM-5.3 GGUF

GGUF quantizations of zai-org/GLM-5.3-BF16, produced from BF16 GGUF weights with an importance matrix.

Quantizations

| Quantization | BPW | Size | Notes |

|---|---:|---:|---|

| Q4_K_M | 4.83 | 454.85 GB | Recommended quality/size balance; importance-matrix calibrated |

| IQ3_XXS | 3.15 | 297.09 GB | Compact 3-bit build; importance-matrix calibrated |

| IQ2_XS | 2.42 | 227.69 GB | Compact 2-bit build; importance-matrix calibrated |

The shared indexer and the full NextN/draft block are kept at Q8_0 where the

calibration matrix has no matching entries. This avoids unsafe low-bit fallback

while retaining importance-matrix mixed quantization for the main model.

Usage

llama-cli -m Q4_K_M/GLM-5.3-Q4_K_M-00001-of-00033.gguf

Use a llama.cpp build with GLM-5.3/DeepSeek-V4 architecture support.

The original model is distributed under the GLM-5.3 license.