RQSA-MIL v1 ONNX classifier for content moderation debuts on HuggingFace
Shivapreetham17 released an ONNX image-classification model for content moderation on HuggingFace, marked not-for-all-audiences with non-standard licensing.
RQSA-MIL v1, a new ONNX image-classification model from Shivapreetham17, targets content moderation workflows. The model card on HuggingFace carries a not-for-all-audiences flag and an "other" license, signaling non-standard usage terms. The ONNX format makes it portable across inference runtimes—ONNX Runtime, TensorRT, and other optimized engines—without framework lock-in, a common choice for production pipelines that need to run classification at scale across heterogeneous hardware: edge devices, cloud GPUs, or CPU-only servers.
The model card does not detail training data, benchmark accuracy, specific content categories the classifier recognizes, parameter count, or input resolution. Usage terms are non-standard and require close review before deployment. Practitioners considering the model will need to run their own validation against known test sets—a common pattern for early-stage moderation classifiers. The release appeared on HuggingFace on July 9, 2026.
