---

base_model: zai-org/GLM-5.3-Flash-BF16

base_model_relation: quantized

library_name: gguf

license: mit

pipeline_tag: text-generation

language:

- en

- zh

tags:

- gguf

- llama.cpp

- glm5_next

- imatrix

---

GLM-5.3-Flash GGUF

GGUF quantizations of zai-org/GLM-5.3-Flash-BF16, converted directly from the official BF16 weights.

Quantizations

| Quantization | Notes |

|---|---|

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

| IQ3_XXS | Mid-size 3.08 BPW build; importance-matrix calibrated |

| IQ2_XS | Compact 2.35 BPW build; importance-matrix calibrated |

The GLM-5.3-sensitive router, shared-expert, mHC, and linear-attention state tensors are preserved at higher precision by the GLM-5.3 llama.cpp quantization rules. The unsupported NextN/MTP draft head is excluded.

Usage

Use a recent GLM-5.3-compatible llama.cpp build and point to the first shard:

llama-cli -m Q4_K_M/GLM-5.3-Flash-Q4_K_M-00001-of-00016.gguf

All shards in the selected quantization directory are required.

License

MIT, inherited from the original model.