This asset contains Qdrant's quantized ONNX export of BAAI/bge-small-en-v1.5. It produces normalized, 384-dimensional English text embeddings on CPU, with inputs truncated to 512 tokens.

These are the same unmodified model and tokenizer files selected by FastEmbed 0.8.0 for TextEmbedding(model_name="BAAI/bge-small-en-v1.5"). This is a raw ONNX custom model, not an MLflow package or a Foundry Local model.

• Source: qdrant/bge-small-en-v1.5-onnx-q.
• Source revision: aa8f8b060edb00e03bfdd08813a2949946c8ba55.
• ONNX file: model_optimized.onnx, 66,465,124 bytes.
• Runtime files total: 67,179,163 bytes, excluding the license and manifest.
• License: MIT; the upstream license is included as LICENSE.

artifact-manifest.json records the size and SHA-256 of every runtime file. Model binaries are stored in the dedicated models container in foundrybenchmarkdatasets, separate from the benchmark datasets in data. The versioned blob prefix includes the source revision; existing artifacts must not be overwritten. Only the asset definitions, license, and manifest are stored in this repository.

Download a pinned registry version during the application build and verify its runtime files against artifact-manifest.json. Registry downloads can preserve the source directory name, so locate the directory containing model_optimized.onnx and its tokenizer files rather than assuming a flat layout:

from pathlib import Path

from fastembed import TextEmbedding

download_root = Path("/app/models/bge-small-en-v1.5-onnx-q")
model_files = list(download_root.rglob("model_optimized.onnx"))
if len(model_files) != 1:
    raise RuntimeError(f"Expected one model_optimized.onnx under {download_root}")
model_dir = model_files[0].parent

model = TextEmbedding(
    model_name="BAAI/bge-small-en-v1.5",
    specific_model_path=str(model_dir),
    local_files_only=True,
    providers=["CPUExecutionProvider"],
)
embeddings = list(model.embed(["Example text"]))

The explicit local directory avoids Hugging Face downloads at runtime. Publishing this asset alone does not change existing consumers; their build and loading configuration must be updated separately.

Version: 1

View in Studio: https://ml.azure.com/registries/azureml/models/bge-small-en-v1.5-onnx-q/version/1