Fix external_data.location: 'model.smoothed.int8.onnx.data' -> 'model.int8.onnx.data' so ORT can find the (renamed) .data file. The .data payload is unchanged.

fix: replace broken vanilla INT8 with SmoothQuant α=0.8 (cos_min 0.64 → 0.99)

init: Octen-Embedding-0.6B SmoothQuant INT8 ONNX (alpha=0.8, cos=~0.92 vs PyTorch ref)

fix: clean re-quantization (per_channel=False, no stale data appending), 1.0GB

Upload README.md with huggingface_hub

Upload model.int8.onnx.data with huggingface_hub

Upload model.int8.onnx with huggingface_hub

Upload folder using huggingface_hub

initial commit