---

license: mit

language:

- en

- zh

pipeline_tag: text-generation

base_model:

- zai-org/GLM-5.2

tags:

- gguf

- glm

- glm-5.2

- abliterated

- quantized

- llama.cpp

---

Huihui GLM-5.2 Abliterated — UD-Q2_K_MXFP4 GGUF

This is a single-quant operational mirror for RunPod cached models. It contains

only the seven UD-Q2_K_MXFP4 GGUF shards from the pinned huihui-ai source.

It does not claim authorship of the model or its quantization.

Destination repository: Jerome0207/Huihui-GLM-5.2-abliterated-UD-Q2_K_MXFP4-GGUF

Provenance

1. zai-org/GLM-5.2
2. huihui-ai abliterated GGUF

Source revision:

994200a058539c553f0977002e58a2f05de845c7.

The seven GGUF shards total 252,600,253,568 bytes. See

checksums.sha256 for the source SHA-256 object identifiers.

Runtime

The repository is intended for a current CUDA build of

llama.cpp. Start with the first shard;

llama.cpp discovers the remaining split files automatically:

GLM-5.2-UD-Q2_K_MXFP4-00001-of-00007.gguf

The validated deployment target uses three 96 GB GPUs, 131,072 context tokens,

one parallel slot and Q8_0 K/V caches.

License and limitations

The source model card declares the MIT license. See NOTICE for attribution

and exact provenance.

This is an “abliterated” derivative with intentionally reduced refusal

behavior. It is not a safety-tuned public service. Deploy it only behind

authentication and rate limits, and run coding/security agents in isolated

sandboxes without production credentials.

GLM-5.2 support in llama.cpp is evolving. Full DSA, Lightning Indexer,

IndexShare and MTP behavior must be validated before treating this deployment

as equivalent to the official inference implementation.