Qwen3-Embedding

Qwen3-Embedding#

The Qwen3-Embedding series is a family of text-embedding models built on the Qwen3 dense transformer backbone. They map text into dense vector representations for semantic search, retrieval, and similarity matching, with strong multilingual coverage.

FuriosaAI publishes the Qwen3-Embedding models under the furiosa-ai organization on the Hugging Face Hub, each shipping a Furiosa Executable Bundle (FXB) for running it on FuriosaAI RNGD with Furiosa-LLM. The same upstream weights also run on other frameworks (such as Sentence Transformers, vLLM, and Transformers); for usage with those, see the upstream model cards linked below.

For the related reranking model see Qwen3-Reranker; for the dense Qwen3 chat models see Qwen3 (dense).

Available Models#

Model

Quantization

RNGD cards

Notes

furiosa-ai/Qwen3-Embedding-0.6B

None (BF16)

1

0.6B text embedding

furiosa-ai/Qwen3-Embedding-4B

None (BF16)

1

4B text embedding

furiosa-ai/Qwen3-Embedding-8B

None (BF16)

1

8B text embedding

  • Architecture: Qwen3 (dense), Qwen3Model

  • Task: Embedding

  • Input / Output: Text / Embeddings (vector)

  • Quantization: No quantization — the models run in their native BF16 precision.

Usage#

To run these models with Furiosa-LLM, follow the examples below after installing Furiosa-LLM and its prerequisites. You can use a model either online through the OpenAI-compatible server or offline through the Furiosa-LLM Python API. Select a variant tab below; the choice is kept in sync across all the examples on this page.

Launch the server#

Serve a model by passing its furiosa-ai/<repo> identifier:

# Launch the server, listening on port 8000 by default
furiosa-llm serve furiosa-ai/Qwen3-Embedding-0.6B

# Launch the server, listening on port 8000 by default
furiosa-llm serve furiosa-ai/Qwen3-Embedding-4B

# Launch the server, listening on port 8000 by default
furiosa-llm serve furiosa-ai/Qwen3-Embedding-8B

When the server is ready, you will see:

INFO:     Started server process [27507]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit)

Basic Usage#

The server exposes an OpenAI-compatible /v1/embeddings endpoint. Request embeddings with curl:

curl http://localhost:8000/v1/embeddings \
    -H "Content-Type: application/json" \
    -d '{
    "model": "furiosa-ai/Qwen3-Embedding-0.6B",
    "input": ["Hello, world!", "How are you?"]
    }' \
    | python -m json.tool

curl http://localhost:8000/v1/embeddings \
    -H "Content-Type: application/json" \
    -d '{
    "model": "furiosa-ai/Qwen3-Embedding-4B",
    "input": ["Hello, world!", "How are you?"]
    }' \
    | python -m json.tool

curl http://localhost:8000/v1/embeddings \
    -H "Content-Type: application/json" \
    -d '{
    "model": "furiosa-ai/Qwen3-Embedding-8B",
    "input": ["Hello, world!", "How are you?"]
    }' \
    | python -m json.tool

Because the endpoint is OpenAI-compatible, you can also use the OpenAI Python client:

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")

response = client.embeddings.create(
    model="furiosa-ai/Qwen3-Embedding-0.6B",
    input=["Hello, world!", "How are you?"],
)

for data in response.data:
    print(f"Index {data.index}: {len(data.embedding)} dimensions")

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")

response = client.embeddings.create(
    model="furiosa-ai/Qwen3-Embedding-4B",
    input=["Hello, world!", "How are you?"],
)

for data in response.data:
    print(f"Index {data.index}: {len(data.embedding)} dimensions")

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")

response = client.embeddings.create(
    model="furiosa-ai/Qwen3-Embedding-8B",
    input=["Hello, world!", "How are you?"],
)

for data in response.data:
    print(f"Index {data.index}: {len(data.embedding)} dimensions")

Advanced Usage#

For offline use, load the model with the LLM constructor (the FXB shipped in the repo is discovered automatically) and call embed to obtain dense vectors:

from furiosa_llm import LLM

with LLM("furiosa-ai/Qwen3-Embedding-0.6B") as llm:
    outputs = llm.embed(["Hello, world!", "How are you?"])
    for output in outputs:
        print(f"{len(output.outputs.embedding)} dimensions")

from furiosa_llm import LLM

with LLM("furiosa-ai/Qwen3-Embedding-4B") as llm:
    outputs = llm.embed(["Hello, world!", "How are you?"])
    for output in outputs:
        print(f"{len(output.outputs.embedding)} dimensions")

from furiosa_llm import LLM

with LLM("furiosa-ai/Qwen3-Embedding-8B") as llm:
    outputs = llm.embed(["Hello, world!", "How are you?"])
    for output in outputs:
        print(f"{len(output.outputs.embedding)} dimensions")

Learn more#