harrier-oss-v1-270m-fastembed

Cleaned ONNX packaging for fastembed.TextEmbedding.

Layout

• config.json
• tokenizer.json
• tokenizer_config.json
• special_tokens_map.json
• model_quantized.onnx
• model_quantized.onnx_data

Notes

• tokenizer_config.json.model_max_length is patched to 32768 to avoid the fastembed truncation overflow that occurs with the Hugging Face sentinel value.
• The model is intended to be loaded with TextEmbedding.add_custom_model(...).
• The target repository for this packaging run is ferrisS/harrier-oss-v1-270m-fastembed.

Example

from fastembed import TextEmbedding
from fastembed.common.model_description import ModelSource, PoolingType

TextEmbedding.add_custom_model(
    model="ferrisS/harrier-oss-v1-270m-fastembed",
    pooling=PoolingType.MEAN,
    normalization=True,
    sources=ModelSource(hf="ferrisS/harrier-oss-v1-270m-fastembed"),
    dim=640,
    model_file="model_quantized.onnx",
    additional_files=["model_quantized.onnx_data"],
)

model = TextEmbedding(model_name="ferrisS/harrier-oss-v1-270m-fastembed")
vectors = list(model.embed(["hello world"]))