---
license: apache-2.0
language:
- en
base_model:
- Qwen/Qwen3-4B
pipeline_tag: text-generation
library_name: transformers
tags:
- text-generation-inference
- moe
---

# **Qwen3-4B-GGUF**

>  Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models. Built upon extensive training, Qwen3 delivers groundbreaking advancements in reasoning, instruction-following, agent capabilities, and multilingual support

## Model Files

| File Name              | Size    | Quantization | Format | Description                      |
| ---------------------- | ------- | ------------ | ------ | -------------------------------- |
| `Qwen3_4B.F32.gguf`    | 16.1 GB | FP32         | GGUF   | Full precision (float32) version |
| `Qwen3_4B.BF16.gguf`   | 8.05 GB | BF16         | GGUF   | BFloat16 precision version       |
| `Qwen3_4B.F16.gguf`    | 8.05 GB | FP16         | GGUF   | Float16 precision version        |
| `Qwen3_4B.Q3_K_M.gguf` | 2.08 GB | Q3\_K\_M     | GGUF   | 3-bit quantized (K M variant)    |
| `Qwen3_4B.Q3_K_S.gguf` | 1.89 GB | Q3\_K\_S     | GGUF   | 3-bit quantized (K S variant)    |
| `Qwen3_4B.Q4_K_M.gguf` | 2.5 GB  | Q4\_K\_M     | GGUF   | 4-bit quantized (K M variant)    |
| `Qwen3_4B.Q4_K_S.gguf` | 2.38 GB | Q4\_K\_S     | GGUF   | 4-bit quantized (K S variant)    |
| `Qwen3_4B.Q5_K_M.gguf` | 2.89 GB | Q5\_K\_M     | GGUF   | 5-bit quantized (K M variant)    |
| `Qwen3_4B.Q8_0.gguf`   | 4.28 GB | Q8\_0        | GGUF   | 8-bit quantized                  |
| `.gitattributes`       | 2.02 kB | —            | —      | Git LFS tracking file            |
| `config.json`          | 31 B    | —            | —      | Configuration placeholder        |
| `README.md`            | 3.6 kB  | —            | —      | Model documentation              |

## Quants Usage 

(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)

| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B.Q2_K.gguf) | Q2_K | 0.4 |  |
| [GGUF](https://huggingface.co/mradermacher/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B.Q3_K_S.gguf) | Q3_K_S | 0.5 |  |
| [GGUF](https://huggingface.co/mradermacher/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B.Q3_K_M.gguf) | Q3_K_M | 0.5 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B.Q3_K_L.gguf) | Q3_K_L | 0.5 |  |
| [GGUF](https://huggingface.co/mradermacher/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B.IQ4_XS.gguf) | IQ4_XS | 0.6 |  |
| [GGUF](https://huggingface.co/mradermacher/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B.Q4_K_S.gguf) | Q4_K_S | 0.6 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B.Q4_K_M.gguf) | Q4_K_M | 0.6 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B.Q5_K_S.gguf) | Q5_K_S | 0.6 |  |
| [GGUF](https://huggingface.co/mradermacher/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B.Q5_K_M.gguf) | Q5_K_M | 0.7 |  |
| [GGUF](https://huggingface.co/mradermacher/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B.Q6_K.gguf) | Q6_K | 0.7 | very good quality |
| [GGUF](https://huggingface.co/mradermacher/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B.Q8_0.gguf) | Q8_0 | 0.9 | fast, best quality |
| [GGUF](https://huggingface.co/mradermacher/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B.f16.gguf) | f16 | 1.6 | 16 bpw, overkill |

Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):

![image.png](https://www.nethype.de/huggingface_embed/quantpplgraph.png)