---
base_model: zai-org/GLM-5.2
base_model_relation: quantized
license: mit
language:
  - en
  - zh
tags:
  - glm
  - moe
  - gguf
  - quantized
  - 4-bit
  - Q4_K_M
  - llama-cpp
pipeline_tag: text-generation
---

# GLM-5.2 Q4_K_M GGUF

First community GGUF quantization of [Z.ai's GLM-5.2](https://huggingface.co/zai-org/GLM-5.2) (745B MoE, 40B active, 1M context).

## Model Details
- **Base model:** [zai-org/GLM-5.2](https://huggingface.co/zai-org/GLM-5.2)
- **Architecture:** glm-dsa (MoE, 256 experts, 79 blocks, Multi-latent Attention, IndexShare)
- **Quantization:** Q4_K_M via llama.cpp (patched for GLM-5.2 MoE layer validation)
- **Format:** Split GGUF (8 shards)
- **Total size:** ~425GB
- **License:** MIT (same as base model)

## Quantization Notes
GLM-5.2 hybrid architecture (3 leading dense blocks + 76 MoE blocks + MTP layers) required a one-line patch to llama.cpp quantizer. Patch: use n_layer_all for the MoE expert tensor bounds check in llama-quant.cpp.

## Produced by
[Molt AI Corp](https://moltaicorp.com)