Qwen3-4B-GGUF/README.md

base_model, license, license_name, license_link, model_creator, model_name, quantized_by, language, pipeline_tag, tags

base_model	license	license_name	license_link	model_creator	model_name	quantized_by	language	pipeline_tag	tags
Qwen/Qwen3-4B	other	qwen-research	https://huggingface.co/Qwen/Qwen3-4B/blob/main/LICENSE	Qwen	Qwen3-4B	Second State Inc.	en	text-generation	chat

---

Qwen3-4B-GGUF

Original Model

Qwen/Qwen3-4B

Run with LlamaEdge

• LlamaEdge version:

  • Thinking: v0.17.0 and above
  • No Thinking: v0.18.2

• Prompt template

  • Prompt type: chatml (for thinking)

  • Prompt string

    <|im_start|>system
    {system_message}<|im_end|>
    <|im_start|>user
    {prompt}<|im_end|>
    <|im_start|>assistant
    

  • Prompt type: qwen3-no-think (for no thinking)

  • Prompt string

    <|im_start|>system
    {system_message}<|im_end|>
    <|im_start|>user
    {user_message_1}<|im_end|>
    <|im_start|>assistant
    {assistant_message_1}<|im_end|>
    <|im_start|>user
    {user_message_2}<|im_end|>
    <|im_start|>assistant
    <think>
    
    </think>
    

• Context size: 128000

• Run as LlamaEdge service

  wasmedge --dir .:. --nn-preload default:GGML:AUTO:Qwen3-4B-Q5_K_M.gguf \
    llama-api-server.wasm \
    --model-name Qwen3-4B \
    --prompt-template chatml \
    --ctx-size 128000
  

• Run as LlamaEdge command app

  wasmedge --dir .:. --nn-preload default:GGML:AUTO:Qwen3-4B-Q5_K_M.gguf \
    llama-chat.wasm \
    --prompt-template chatml \
    --ctx-size 128000
  

Quantized GGUF Models

Name	Quant method	Bits	Size	Use case
Qwen3-4B-Q2_K.gguf	Q2_K	2	1.67 GB	smallest, significant quality loss - not recommended for most purposes
Qwen3-4B-Q3_K_L.gguf	Q3_K_L	3	2.24 GB	small, substantial quality loss
Qwen3-4B-Q3_K_M.gguf	Q3_K_M	3	2.08 GB	very small, high quality loss
Qwen3-4B-Q3_K_S.gguf	Q3_K_S	3	1.89 GB	very small, high quality loss
Qwen3-4B-Q4_0.gguf	Q4_0	4	2.37 GB	legacy; small, very high quality loss - prefer using Q3_K_M
Qwen3-4B-Q4_K_M.gguf	Q4_K_M	4	2.50 GB	medium, balanced quality - recommended
Qwen3-4B-Q4_K_S.gguf	Q4_K_S	4	2.38 GB	small, greater quality loss
Qwen3-4B-Q5_0.gguf	Q5_0	5	2.82 GB	legacy; medium, balanced quality - prefer using Q4_K_M
Qwen3-4B-Q5_K_M.gguf	Q5_K_M	5	2.89 GB	large, very low quality loss - recommended
Qwen3-4B-Q5_K_S.gguf	Q5_K_S	5	2.82 GB	large, low quality loss - recommended
Qwen3-4B-Q6_K.gguf	Q6_K	6	3.31 GB	very large, extremely low quality loss
Qwen3-4B-Q8_0.gguf	Q8_0	8	4.28 GB	very large, extremely low quality loss - not recommended
Qwen3-4B-f16.gguf	f16	16	8.05 GB

Quantized with llama.cpp b5097