---
license: mit
base_model:
  - inclusionAI/Ling-3.0-tiny
pipeline_tag: text-generation
tags:
  - gguf
  - bailingmoe3
  - mixture-of-experts
  - conversational
---

# Ling-3.0-tiny GGUF

GGUF conversions of [inclusionAI/Ling-3.0-tiny](https://huggingface.co/inclusionAI/Ling-3.0-tiny), converted directly from the released BF16 safetensors.

## Compatibility

Ling-3.0-tiny uses the new `bailingmoe3` GGUF architecture and its Q-LoRA attention path. Until support is merged upstream, use the following llama.cpp branch:

https://github.com/aetherbird/llama.cpp/tree/bailingmoe3-support

Stock llama.cpp builds without BailingMoE3 support will not load these files.

Upstream PR: https://github.com/ggml-org/llama.cpp/pull/26608

## Files

| Quant | Size |
| --- | ---: |
| BF16 | 15.8 GB |
| UD-Q8_K_XL | 11.19 GB |
| Q8_0 | 8.41 GB |
| UD-Q6_K_XL | 7.27 GB |
| Q6_K | 6.50 GB |
| Q5_K_M | 5.64 GB |
| Q5_K_S | 5.48 GB |
| Q4_K_M | 4.82 GB |
| MXFP4_MOE | 4.72 GB |
| Q3_K_M | 3.84 GB |
| IQ2_M | 2.70 GB |
| Q1_0 | 1.30 GB |

## Importance Matrix

IQ2_M was generated with a model-specific importance matrix:

- Wikitext-2 raw training text
- 100 chunks
- 512 tokens per chunk
- 51,200 calibration tokens total
- 332 matrix entries

## XL Quantization Recipes

`UD-Q8_K_XL` uses Q8_0 for the main expert gate and up tensors. Token embeddings, expert down projections, attention and Q-LoRA projections, and KDA projections remain BF16.

`UD-Q6_K_XL` uses Q6_K for the main expert gate and up tensors. Token embeddings, output weights, expert down projections, attention and Q-LoRA projections, and KDA projections use Q8_0. It was generated with the importance matrix described above.

## Architecture

- 7.9B total parameters and 1.3B active parameters per token
- 24 layers: 18 KDA layers and 6 MLA layers
- 128 routed experts, 8 active per token, plus 1 shared expert
- Q-LoRA rank 256 and KV-LoRA rank 512
- 131,072-token context in the released configuration
- No bundled MTP block (`num_nextn_predict_layers: 0`)

## Validation

- BF16 conversion completed with 526 tensors, including all 18 Q-LoRA tensors
- CPU and CUDA architecture tests passed
- BF16, Q8_0, Q6_K, Q4_K_M, and MXFP4_MOE loaded and generated tokens with CUDA
- Q1_0, IQ2_M, Q3_K_M, Q5_K_S, and Q5_K_M passed CPU-only prompt processing and token generation tests
- UD-Q6_K_XL and UD-Q8_K_XL passed CPU-only prompt processing and token generation tests
- CUDA testing used an RTX 4070 and RTX 3060

## Build

```bash
git clone --branch bailingmoe3-support https://github.com/aetherbird/llama.cpp.git
cd llama.cpp
cmake -B build -DGGML_CUDA=ON
cmake --build build --config Release -j --target llama-cli llama-server
```

## Usage

```bash
./build/bin/llama-server \
  -m Ling-3.0-tiny-Q4_K_M.gguf \
  -c 131072 \
  -ngl auto \
  --flash-attn auto
```

Thinking is enabled by default. Recommended sampling parameters from the source model card are `temperature=1.0`, `top_p=0.95`, and `top_k=20`.