Modelpublic

techtheist/deberta-v3-small-tasksource-nli-onnx

sourceHugging Faceapache-2.0updated 2mo agoView on Hugging Face

0likes62downloads

Model Card

deberta-v3-small-tasksource-nli — ONNX (int8)

ONNX export of sileod/deberta-v3-small-tasksource-nli (Apache-2.0), dynamically quantized to uint8 weights with onnxruntime.

Exported with optimum-cli export onnx --task text-classification, then onnxruntime.quantization.quantize_dynamic(weight_type=QUInt8).

Made for Engram Alpha's local contradiction layer (entailment / neutral / contradiction over sentence pairs, no LLM). All credit for the model to @sileod / tasksource.