[Doc]: 请问支持的模型里，有没有支持嵌入维度可设置为1536的embedding模型？

[winni0]

📚 The doc issue

请问支持的模型里，有没有支持嵌入维度可设置为1536的embedding模型？

Suggest a potential alternative/fix

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[noooop]

refer to mteb/leaderboard，https://huggingface.co/spaces/mteb/leaderboard

There are not many embedding models with an embedding dimension is 1536. Most open-source models have dimensions <=1024

you might want to consider models supporting Matryoshka.

for example, Qwen3-Embedding-4B defaults to a dimension of 2560, You can get a 1536-dimensional vector by setting dimensions.

curl http://127.0.0.1:8000/v1/embeddings \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
    "input": "Follow the white rabbit.",
    "model": "Qwen3-Embedding-4B",
    "encoding_format": "float",
    "dimensions": 1536
  }'

More information https://docs.vllm.ai/en/latest/models/pooling_models.html#matryoshka-embeddings

[winni0]

参考mteb/leaderboard，https://huggingface.co/spaces/mteb/leaderboard

嵌入维度为 1536 的嵌入模型并不多。大多数开源模型的维度 <=1024

您可能需要考虑支持 Matryoshka 的模型。

例如Qwen3-Embedding-4B默认维度为2560，通过设置维度可以得到1536维的向量。

curl http://127.0.0.1:8000/v1/embeddings \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
    "input": "Follow the white rabbit.",
    "model": "Qwen3-Embedding-4B",
    "encoding_format": "float",
    "dimensions": 1536
  }'

更多信息https://docs.vllm.ai/en/latest/models/pooling_models.html#matryoshka-embeddings

使用CUDA_VISIBLE_DEVICES=2 vllm serve /znfs/suanfa/models/Qwen3-Embedding-4B --task embed --dtype auto --host 0.0.0.0 --port 10014或者CUDA_VISIBLE_DEVICES=2 vllm serve /znfs/suanfa/models/Qwen3-Embedding-4B --task embed --dtype auto --host 0.0.0.0 --port 10014 --hf_overrides '{"matryoshka_dimensions":[1536]}'命令启动服务。
使用以下命令进行测试：
curl http://127.0.0.1:10014/v1/embeddings -H 'accept: application/json' -H 'Content-Type: application/json' -d '{
"input": "水木清华侯君来",
"model": "/znfs/suanfa/models/Qwen3-Embedding-4B",
"encoding_format": "float",
"dimensions": 1536
}'。
报错：

[noooop]

需要vllm >= v0.8.4 才支持 matryoshka_dimensions

PTAL #16331

[dengcao]

You can use Vllm to deploy the Qwen3-Embedding&Qwen3-Reranker large model.There is already a temporary solution:
以下是通过vllm快捷安装Qwen3-Embedding-0.6B/4B/8B，Qwen3-Reranker-0.6B/4B/8B的办法。

Docker 快速部署（一键安装包）：
Docker Compose Quick Installation:

Hugging Face：
https://huggingface.co/dengcao/Qwen3-Embedding-0.6B
https://huggingface.co/dengcao/Qwen3-Embedding-4B
https://huggingface.co/dengcao/Qwen3-Embedding-8B
https://huggingface.co/dengcao/Qwen3-Reranker-0.6B
https://huggingface.co/dengcao/Qwen3-Reranker-4B
https://huggingface.co/dengcao/Qwen3-Reranker-8B

ModelScope：
https://www.modelscope.cn/models/dengcao/Qwen3-Embedding-0.6B
https://www.modelscope.cn/models/dengcao/Qwen3-Embedding-4B
https://www.modelscope.cn/models/dengcao/Qwen3-Embedding-8B
https://www.modelscope.cn/models/dengcao/Qwen3-Reranker-0.6B
https://www.modelscope.cn/models/dengcao/Qwen3-Reranker-4B
https://www.modelscope.cn/models/dengcao/Qwen3-Reranker-8B

GitHub：
https://github.com/dengcao/Qwen3-Reranker-8B
https://github.com/dengcao/Qwen3-Reranker-4B
https://github.com/dengcao/Qwen3-Reranker-0.6B

[youngerstyle]

I also searched for a long time, but finally found one, I hope it will be helpful to you: Alibaba-NLP/gte-Qwen2-1.5B-instruct