Qwen

/

Qwen3-Embedding-0.6B

like 738

Follow

Qwen 57.7k

It is truncation, not pooling.

like this：https://github.com/QwenLM/Qwen3-Embedding/issues/2#issuecomment-2948660134

Use truncate_dim=x when creating the model:

model = SentenceTransformer("Qwen/Qwen3-Embedding-0.6B", truncate_dim=512)

The is_matryoshka parameter is missing from config.json. You need to start vllm with hf_overrides={"is_matryoshka": True}.
Reference: vLLM docs