修复虚拟环境创建时qwen-vl-utils版本错误导致的Qwen3-Vl-Embedding-8B向量化返回同一个向量的问题

[yuyue0823]

System Info / 系統信息

Cuda：12.4
NVIDIA L20

Running Xinference with Docker? / 是否使用 Docker 运行 Xinfernece？

• docker / docker
• pip install / 通过 pip install 安装
• installation from source / 从源码安装

Version info / 版本信息

2.8.0

The command used to start Xinference / 用以启动 xinference 的命令

默认运行命令

Reproduction / 复现过程

1、v1/embeddings调用，参数"model":"Qwen3-Vl-Embedding-8B","task":"image","input":[""]
2、docker logs -f 容器名，报错got an unexpected keyword argument 'imaghe_patch_size'

Expected behavior / 期待表现

修复一下这个导包导致的报错，原因是在创建虚拟环境的时候，会读取xinference/model/embedding/model_spec.json下的配置文件，文件内容为:
{
"version": 2,
"model_name": "Qwen3-VL-Embedding-8B",
"dimensions": 4096,
"max_tokens": 8192,
"language": [
"zh",
"en"
],
"model_specs": [
{
"model_format": "pytorch",
"model_src": {
"huggingface": {
"model_id": "Qwen/Qwen3-VL-Embedding-8B",
"model_revision": "a12d6118f720ceb6d95f7d1cad4e8aeccddd9340",
"quantizations": [
"none"
]
},
"modelscope": {
"model_id": "Qwen/Qwen3-VL-Embedding-8B",
"quantizations": [
"none"
]
}
}
}
],
"virtualenv": {
"packages": [
"transformers==4.57.1 ; #engine# == "sentence_transformers"",
"qwen-vl-utils==0.0.11",
"Pillow",
"#vllm_dependencies# ; #engine# == "vllm"",
"#system_numpy# ; #engine# == "vllm""
]
},
"featured": false,
"updated_at": 1769418314
}

请将这里的 "qwen-vl-utils==0.0.11"，修改为 "qwen-vl-utils>=0.0.14"

[la1ty]

你 Fork 之后改掉，然后提 Pull Request，这样可能会快一点（划掉