n24q02m/Qwen3-Embedding-0.6B-GGUF

GGUF-quantized version of Qwen/Qwen3-Embedding-0.6B for use with qwen3-embed and llama-cpp-python.

Available Variants

Variant	File	Size	Description
Q4_K_M	qwen3-embedding-0.6b-q4-k-m.gguf	378 MB	4-bit quantization (recommended)

Usage

qwen3-embed

pip install qwen3-embed[gguf]

from qwen3_embed import TextEmbedding

model = TextEmbedding("n24q02m/Qwen3-Embedding-0.6B-GGUF")
embeddings = list(model.embed(["Hello world"]))               # 1024-dim

# MRL: reduce dimension
embeddings_256 = list(model.embed(["Hello world"], dim=256))  # 256-dim

# Query with instruction
query_emb = list(model.query_embed("What is machine learning?"))

llama-cpp-python (direct)

from llama_cpp import Llama

model = Llama(
    model_path="qwen3-embedding-0.6b-q4-k-m.gguf",
    embedding=True,
    pooling_type=3,  # LLAMA_POOLING_TYPE_LAST
    n_ctx=32768,
)
result = model.create_embedding("Hello world")

Conversion Details

• Source: Qwen/Qwen3-Embedding-0.6B
• Method: convert_hf_to_gguf.py (F16) + llama-quantize (Q4_K_M)

Related

• ONNX variants: n24q02m/Qwen3-Embedding-0.6B-ONNX