---

license: apache-2.0

base_model:

• Nanbeige/Nanbeige4.2-3B

base_model_relation: quantized

library_name: gguf

pipeline_tag: text-generation

tags:

• wekala
• gguf

---

nanbeige4.2-3b-gguf

Consumer-supplied Wekala model-library assets derived from the attributed official publisher. Models and this repository are distributed separately from the Wekala SDK.

Origin and licence

| Field | Value |

| --- | --- |

| Original developer | Nanbeige |

| Original model | Nanbeige/Nanbeige4.2-3B |

| Pinned source revision | b82e54bd609793562a75cbf9337970a93369eab5 |

| Original publication | 2026-07 (month) |

| Date evidence | Publisher source |

| Licence | apache-2.0; preserved in LICENSE |

| Full source parameters | 4,169,800,704 (includes embeddings and source modalities) |

| Nile Academy role | Official artifact mirror or explicitly recorded format conversion; no additional training |

Original publisher cards and available notices are preserved in provenance/. The original model licence governs the weights; this repository grants no rights to the Wekala SDK.

Files and quantization

Q8_0 is the conservative higher-precision option. Four calibration-text chunks showed PPL increases of 0.53% (Q8), 2.97% (Q6), 3.31% (Q5) and 16.60% (Q4) versus BF16. This is not held-out task evaluation: Q4 is experimental, and no general quality-loss or 8GB-fit qualification is claimed.

| File | Quantization | Size (GB) | Origin |

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

| Nanbeige4.2-3B-Q8_0-llamacpp-4413a19.gguf | Q8_0 | 4.435 | Wekala conversion from official BF16 |

| Nanbeige4.2-3B-Q6_K-llamacpp-4413a19.gguf | Q6_K | 3.425 | Wekala conversion from official BF16 |

| Nanbeige4.2-3B-Q5_K_M-llamacpp-4413a19.gguf | Q5_K_M | 2.987 | Wekala conversion from official BF16 |

| Nanbeige4.2-3B-Q4_K_M-llamacpp-4413a19.gguf | Q4_K_M | 2.575 | Wekala conversion from official BF16 |

Exact file SHA-256 values, upstream paths/revisions and conversion receipts are recorded in model-metadata.json.

Published benchmarks

Publisher-reported source-model scores follow. They are not Wekala measurements of these quantized files. Harness, precision, reasoning settings and task versions differ; do not treat them as a normalized cross-model ranking.

| Benchmark | Score (%) | Mode / settings |

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

| GPQA Diamond | 87.4 | reasoning; Publisher source protocol; not normalized across models or measured on this quantization. |

| LiveCodeBench v6 | 72.5 | reasoning; Publisher source protocol; not normalized across models or measured on this quantization. |

| SWE-bench Verified | 63.6 | publisher agent harness; Publisher source protocol; not normalized across models or measured on this quantization. |

| Terminal-Bench 2.0 | 44.1 | publisher agent harness; Publisher source protocol; not normalized across models or measured on this quantization. |

Exact source card.

Qualification and use

BF16/Q8/Q6/Q5/Q4 passed Windows CPU raw-text load/numerical screens. Chat/tool/template fidelity, sustained GPU fit and mobile/browser qualification remain pending.

Wekala broad quantization-quality qualification: not qualified. No general “no significant quality loss” claim is made. Official QAT or publisher comparisons are attributed separately. Before recommendation, use paired reference/variant evaluation across instruction, knowledge, math, multilingual/Arabic and tool tasks under the registered Wekala campaign.

Device fit is separate: weight size excludes context state, scratch buffers and driver headroom. CPU and CUDA are execution backends for the same weight file, not separate models. Mobile/browser and sustained 8GB-VRAM operation require their own measured profiles.

Models can produce incorrect or unsafe outputs. Tool calls are proposals subject to the consuming application’s permissions and controller admission.