metadata

base_model: Qwen/Qwen2.5-VL-32B-Instruct
license: apache-2.0
model_creator: Qwen
model_name: Qwen2.5-VL-32B-Instruct
quantized_by: Second State Inc.
language:
  - en
pipeline_tag: image-text-to-text
tags:
  - multimodal
library_name: transformers

Qwen2.5-VL-32B-Instruct-GGUF

Original Model

Qwen/Qwen2.5-VL-32B-Instruct

Run with LlamaEdge

LlamaEdge version: v0.18.4

Prompt template

Prompt type: qwen2-vision

Prompt string

<|im_start|>system
{system_prompt}<|im_end|>
<|im_start|>user
<|vision_start|>{image_placeholder}<|vision_end|>{user_prompt}<|im_end|>
<|im_start|>assistant

Context size: 32000

Run as LlamaEdge service

wasmedge --dir .:. \
  --nn-preload default:GGML:AUTO:Qwen2.5-VL-32B-Instruct-Q5_K_M.gguf \
  llama-api-server.wasm \
  --model-name Qwen2.5-VL-32B-Instruct \
  --prompt-template qwen2-vision \
  --llava-mmproj Qwen2.5-VL-32B-Instruct-vision.gguf \
  --ctx-size 32000

Quantized GGUF Models

Name	Quant method	Bits	Size	Use case
Qwen2.5-VL-32B-Instruct-Q2_K.gguf	Q2_K	2	12.3 GB	smallest, significant quality loss - not recommended for most purposes
Qwen2.5-VL-32B-Instruct-Q3_K_L.gguf	Q3_K_L	3	17.2 GB	small, substantial quality loss
Qwen2.5-VL-32B-Instruct-Q3_K_M.gguf	Q3_K_M	3	15.9 GB	very small, high quality loss
Qwen2.5-VL-32B-Instruct-Q3_K_S.gguf	Q3_K_S	3	14.4 GB	very small, high quality loss
Qwen2.5-VL-32B-Instruct-Q4_0.gguf	Q4_0	4	18.6 GB	legacy; small, very high quality loss - prefer using Q3_K_M
Qwen2.5-VL-32B-Instruct-Q4_K_M.gguf	Q4_K_M	4	19.9 GB	medium, balanced quality - recommended
Qwen2.5-VL-32B-Instruct-Q4_K_S.gguf	Q4_K_S	4	18.8 GB	small, greater quality loss
Qwen2.5-VL-32B-Instruct-Q5_0.gguf	Q5_0	5	22.6 GB	legacy; medium, balanced quality - prefer using Q4_K_M
Qwen2.5-VL-32B-Instruct-Q5_K_M.gguf	Q5_K_M	5	23.3 GB	large, very low quality loss - recommended
Qwen2.5-VL-32B-Instruct-Q5_K_S.gguf	Q5_K_S	5	22.6 GB	large, low quality loss - recommended
Qwen2.5-VL-32B-Instruct-Q6_K.gguf	Q6_K	6	26.9 GB	very large, extremely low quality loss
Qwen2.5-VL-32B-Instruct-Q8_0.gguf	Q8_0	8	34.8 GB	very large, extremely low quality loss - not recommended
Qwen2.5-VL-32B-Instruct-f16-00001-of-00003.gguf	f16	16	29.8 GB
Qwen2.5-VL-32B-Instruct-f16-00002-of-00003.gguf	f16	16	29.8 GB
Qwen2.5-VL-32B-Instruct-f16-00003-of-00003.gguf	f16	16	5.87 GB
Qwen2.5-VL-32B-Instruct-vision.gguf	f16	16	1.38 GB

Quantized with llama.cpp b5196