Qwen3-Embedding-4B-fp16-ONNX

NOTE!

This model was exported to ONNX from Qwen/Qwen3-Embedding-4B for use with Vespa's HuggingFaceEmbedder using the export_qwen_onnx_with_pooling_and_no_position_ids.py attached script. The original model can be found here

Why ONNX?

Vespa's HuggingFaceEmbedder requires ONNX format and does not natively support decoder-only embedding models. Decoder models like Qwen3-Embedding use last-token pooling — only the final non-padding token attends to the full sequence and carries the sentence representation. Vespa's built-in pooling options (mean, cls, none) don't include last-token pooling, so it must be baked into the model itself.

What was changed during export

1. position_ids removed from inputs — Vespa's HuggingFaceEmbedder does not pass position_ids at inference time. The model is exported without this input so it is compatible out of the box.
   • imoportant: wihtout this, when exporting onnx we loose information about optionallity of arguments and so it becomes a required input! so this trick overcomes this.
2. Last-token pooling baked in — the wrapper module computes the last non-padding token position from the attention mask and extracts that token's hidden state, replacing the need for a separate pooling step.
3. Output shape (batch, 1, hidden) — the pooled vector is unsqueezed to a sequence dimension of 1. This makes it compatible with both Vespa's mean pooling strategy (averaging over a single element is a no-op) and SentenceTransformer's Pooling module with pooling_mode_lasttoken.
4. SentenceTransformer config files included — modules.json, 1_Pooling/config.json (last-token mode), and 2_Normalize/config.json are bundled so the model can be loaded directly with SentenceTransformer(..., backend="onnx"). Normalization is not baked into the ONNX model — it is applied by Vespa and the ST Normalize module at runtime.

---

Qwen3-Embedding-4B

<p align="center"> <img src="https://qianwen-res.oss-accelerate-overseas.aliyuncs.com/logo_qwen3.png" width="400"/> <p>

Highlights

The Qwen3 Embedding model series is the latest proprietary model of the Qwen family, specifically designed for text embedding and ranking tasks. Building upon the dense foundational models of the Qwen3 series, it provides a comprehensive range of text embeddings and reranking models in various sizes (0.6B, 4B, and 8B). This series inherits the exceptional multilingual capabilities, long-text understanding, and reasoning skills of its foundational model. The Qwen3 Embedding series represents significant advancements in multiple text embedding and ranking tasks, including text retrieval, code retrieval, text classification, text clustering, and bitext mining.

Exceptional Versatility: The embedding model has achieved state-of-the-art performance across a wide range of downstream application evaluations. The 8B size embedding model ranks No.1 in the MTEB multilingual leaderboard (as of June 5, 2025, score 70.58), while the reranking model excels in various text retrieval scenarios.

Comprehensive Flexibility: The Qwen3 Embedding series offers a full spectrum of sizes (from 0.6B to 8B) for both embedding and reranking models, catering to diverse use cases that prioritize efficiency and effectiveness. Developers can seamlessly combine these two modules. Additionally, the embedding model allows for flexible vector definitions across all dimensions, and both embedding and reranking models support user-defined instructions to enhance performance for specific tasks, languages, or scenarios.

Multilingual Capability: The Qwen3 Embedding series offer support for over 100 languages, thanks to the multilingual capabilites of Qwen3 models. This includes various programming languages, and provides robust multilingual, cross-lingual, and code retrieval capabilities.

Model Overview

Qwen3-Embedding-4B has the following features:

• Model Type: Text Embedding
• Supported Languages: 100+ Languages
• Number of Paramaters: 4B
• Context Length: 32k
• Embedding Dimension: Up to 2560, supports user-defined output dimensions ranging from 32 to 2560

For more details, including benchmark evaluation, hardware requirements, and inference performance, please refer to our blog, GitHub.

Qwen3 Embedding Series Model list

Note:

• MRL Support indicates whether the embedding model supports custom dimensions for the final embedding.
• Instruction Aware notes whether the embedding or reranking model supports customizing the input instruction according to different tasks.
• Our evaluation indicates that, for most downstream tasks, using instructions (instruct) typically yields an improvement of 1% to 5% compared to not using them. Therefore, we recommend that developers create tailored instructions specific to their tasks and scenarios. In multilingual contexts, we also advise users to write their instructions in English, as most instructions utilized during the model training process were originally written in English.

Usage

With Transformers versions earlier than 4.51.0, you may encounter the following error:

KeyError: 'qwen3'

Sentence Transformers Usage

# Requires transformers>=4.51.0
# Requires sentence-transformers>=2.7.0

from sentence_transformers import SentenceTransformer

# Load the model
model = SentenceTransformer("Qwen/Qwen3-Embedding-4B")

# We recommend enabling flash_attention_2 for better acceleration and memory saving,
# together with setting `padding_side` to "left":
# model = SentenceTransformer(
#     "Qwen/Qwen3-Embedding-4B",
#     model_kwargs={"attn_implementation": "flash_attention_2", "device_map": "auto"},
#     tokenizer_kwargs={"padding_side": "left"},
# )

# The queries and documents to embed
queries = [
    "What is the capital of China?",
    "Explain gravity",
]
documents = [
    "The capital of China is Beijing.",
    "Gravity is a force that attracts two bodies towards each other. It gives weight to physical objects and is responsible for the movement of planets around the sun.",
]

# Encode the queries and documents. Note that queries benefit from using a prompt
# Here we use the prompt called "query" stored under `model.prompts`, but you can
# also pass your own prompt via the `prompt` argument
query_embeddings = model.encode(queries, prompt_name="query")
document_embeddings = model.encode(documents)

# Compute the (cosine) similarity between the query and document embeddings
similarity = model.similarity(query_embeddings, document_embeddings)
print(similarity)
# tensor([[0.7534, 0.1147],
#         [0.0320, 0.6258]])

Transformers Usage

# Requires transformers>=4.51.0
import torch
import torch.nn.functional as F

from torch import Tensor
from transformers import AutoTokenizer, AutoModel


def last_token_pool(last_hidden_states: Tensor,
                 attention_mask: Tensor) -> Tensor:
    left_padding = (attention_mask[:, -1].sum() == attention_mask.shape[0])
    if left_padding:
        return last_hidden_states[:, -1]
    else:
        sequence_lengths = attention_mask.sum(dim=1) - 1
        batch_size = last_hidden_states.shape[0]
        return last_hidden_states[torch.arange(batch_size, device=last_hidden_states.device), sequence_lengths]


def get_detailed_instruct(task_description: str, query: str) -> str:
    return f'Instruct: {task_description}\nQuery:{query}'

# Each query must come with a one-sentence instruction that describes the task
task = 'Given a web search query, retrieve relevant passages that answer the query'

queries = [
    get_detailed_instruct(task, 'What is the capital of China?'),
    get_detailed_instruct(task, 'Explain gravity')
]
# No need to add instruction for retrieval documents
documents = [
    "The capital of China is Beijing.",
    "Gravity is a force that attracts two bodies towards each other. It gives weight to physical objects and is responsible for the movement of planets around the sun."
]
input_texts = queries + documents

tokenizer = AutoTokenizer.from_pretrained('Qwen/Qwen3-Embedding-4B', padding_side='left')
model = AutoModel.from_pretrained('Qwen/Qwen3-Embedding-4B')

# We recommend enabling flash_attention_2 for better acceleration and memory saving.
# model = AutoModel.from_pretrained('Qwen/Qwen3-Embedding-4B', attn_implementation="flash_attention_2", torch_dtype=torch.float16).cuda()

max_length = 8192

# Tokenize the input texts
batch_dict = tokenizer(
    input_texts,
    padding=True,
    truncation=True,
    max_length=max_length,
    return_tensors="pt",
)
batch_dict.to(model.device)
outputs = model(**batch_dict)
embeddings = last_token_pool(outputs.last_hidden_state, batch_dict['attention_mask'])

# normalize embeddings
embeddings = F.normalize(embeddings, p=2, dim=1)
scores = (embeddings[:2] @ embeddings[2:].T)
print(scores.tolist())
# [[0.7534257769584656, 0.1146894246339798], [0.03198453038930893, 0.6258305311203003]]

vLLM Usage

# Requires vllm>=0.8.5
import torch
import vllm
from vllm import LLM

def get_detailed_instruct(task_description: str, query: str) -> str:
    return f'Instruct: {task_description}\nQuery:{query}'

# Each query must come with a one-sentence instruction that describes the task
task = 'Given a web search query, retrieve relevant passages that answer the query'

queries = [
    get_detailed_instruct(task, 'What is the capital of China?'),
    get_detailed_instruct(task, 'Explain gravity')
]
# No need to add instruction for retrieval documents
documents = [
    "The capital of China is Beijing.",
    "Gravity is a force that attracts two bodies towards each other. It gives weight to physical objects and is responsible for the movement of planets around the sun."
]
input_texts = queries + documents

model = LLM(model="Qwen/Qwen3-Embedding-4B", task="embed")

outputs = model.embed(input_texts)
embeddings = torch.tensor([o.outputs.embedding for o in outputs])
scores = (embeddings[:2] @ embeddings[2:].T)
print(scores.tolist())
# [[0.7525103688240051, 0.1143278032541275], [0.030893627554178238, 0.6239761114120483]]

📌 Tip: We recommend that developers customize the instruct according to their specific scenarios, tasks, and languages. Our tests have shown that in most retrieval scenarios, not using an instruct on the query side can lead to a drop in retrieval performance by approximately 1% to 5%.

Text Embeddings Inference (TEI) Usage

You can either run / deploy TEI on NVIDIA GPUs as:

docker run --gpus all -p 8080:80 -v hf_cache:/data --pull always ghcr.io/huggingface/text-embeddings-inference:1.7.2 --model-id Qwen/Qwen3-Embedding-4B --dtype float16

Or on CPU devices as:

docker run -p 8080:80 -v hf_cache:/data --pull always ghcr.io/huggingface/text-embeddings-inference:cpu-1.7.2 --model-id Qwen/Qwen3-Embedding-4B --dtype float16

And then, generate the embeddings sending a HTTP POST request as:

curl http://localhost:8080/embed \
    -X POST \
    -d '{"inputs": ["Instruct: Given a web search query, retrieve relevant passages that answer the query\nQuery: What is the capital of China?", "Instruct: Given a web search query, retrieve relevant passages that answer the query\nQuery: Explain gravity"]}' \
    -H "Content-Type: application/json"

Evaluation

MTEB (Multilingual)

Note: For compared models, the scores are retrieved from MTEB online leaderboard on May 24th, 2025.

MTEB (Eng v2)

C-MTEB (MTEB Chinese)

Citation

If you find our work helpful, feel free to give us a cite.

@article{qwen3embedding,
  title={Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models},
  author={Zhang, Yanzhao and Li, Mingxin and Long, Dingkun and Zhang, Xin and Lin, Huan and Yang, Baosong and Xie, Pengjun and Yang, An and Liu, Dayiheng and Lin, Junyang and Huang, Fei and Zhou, Jingren},
  journal={arXiv preprint arXiv:2506.05176},
  year={2025}
}