---

license: other

license_name: openpangu-license

license_link: https://huggingface.co/openpangu/openPangu-2.0-Flash/blob/main/LICENSE

base_model:

• openpangu/openPangu-2.0-Flash

pipeline_tag: text-generation

tags:

• gguf
• ik_llama
• openpangu
• moe
• mla
• iq4-nl
• imatrix

---

openPangu-2.0-Flash ik_llama GGUF

GGUF quantizations of openPangu-2.0-Flash, Huawei's 92B-A6B MoE trained to 512K context, converted for ik_llama.cpp.

The OpenPangu architecture merged into ik_llama.cpp through PR #2065. These are self-contained conversions with the DSA/SWA schedule and indexer configuration stored in GGUF metadata.

Files

| file | quant | size | notes |

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

| openPangu-2.0-Flash-IQ4_NL-imatrix-mixed-v1.gguf | IQ4_NL mixed | 53.46 GiB | Importance-matrix quantization for eligible base-model matrices, with higher precision retained for fragile and uncalibrated tensor families |

| openPangu-2.0-Flash-ik_llama-Q4_K_M.gguf | Q4_K_M | 56.7 GiB | Earlier clean conversion using explicit high-precision overrides |

The IQ4_NL file is 3.25 GiB, or 5.7%, smaller than the Q4_K_M control.

IQ4_NL quantization

The importance matrix was collected from 100 independent 512-token chunks drawn from mixed Chinese and English documentation, code, and general prose. Its 700 entries cover all 44 base MoE blocks. The standalone collector does not exercise the three NextN/MTP blocks, so blocks 46 through 48 use the reviewed Q4_K_M tensor policy instead of uncalibrated importance data.

The final GGUF contains 1,681 tensors:

| type | tensors |

|---|---:|

| F32 | 800 |

| F16 | 288 |

| IQ4_NL | 363 |

| Q8_0 | 98 |

| Q4_K | 76 |

| Q5_K | 49 |

| Q6_K | 7 |

MoME convolution weights, mHC tensors, attention sinks, and DSA indexer weights remain at their reviewed higher-precision types. Absorbed MLA K-B tensors use Q8_0 and V-B tensors use Q4_K. The quantization source was the BF16 conversion, not the existing Q4_K_M file.

SHA-256:

57c828d080ce81603d348a8409f574472ac24ee502f82cecb51b488b23e570c9  openPangu-2.0-Flash-IQ4_NL-imatrix-mixed-v1.gguf

Usage

CPU:

llama-server \
  -m openPangu-2.0-Flash-IQ4_NL-imatrix-mixed-v1.gguf \
  -c 20480 --jinja

Basic CUDA placement, with routed experts in system RAM:

llama-server \
  -m openPangu-2.0-Flash-IQ4_NL-imatrix-mixed-v1.gguf \
  -c 20480 --jinja \
  -ngl 999 -ot exps=CPU -fa off -ctk q8_0

--jinja is required because the embedded chat template carries the model's message format. The template enables thinking under the kwarg name thinking; pass chat_template_kwargs: {"thinking": false} in a request to disable it.

NextN/MTP speculation is optional, up to heads=3.

Q4_K_M conversion notes

The Q4_K_M file uses --custom-q overrides for convolution, mHC, sink, and indexer tensors. The k=3 MoME convolution tensors are not block-quantization compatible, while the other preserved families cost little relative to the full model.

Its SHA-256 is:

1fd980491573a39fcaf4a08acfaf2b15a6ef606cc947998d4f23a8b3e4f7a79c  openPangu-2.0-Flash-ik_llama-Q4_K_M.gguf

License

This is a conversion of Huawei's openPangu-2.0-Flash weights and inherits the openPangu License, which includes an attribution ("Powered by openPangu") notice requirement and a European Union use restriction. Read the license before deploying.

Powered by openPangu.