independently-platform/picoguardian-v3

docs: fix model-index YAML schema (add dataset field per entry)

chore(v3): remove superseded ONNX INT8 artefact (constant-output on 4060)

chore(v3): shipping torch-fp16 checkpoint + use-case calibration + model card

Upload folder using huggingface_hub

picoguardian-v3: remove Memcpy bake-ins (use_gpu=False fuse) + freeze seq=1024

picoguardian-v3: INT8 weight-only (MatMulNBits bits=8), block_size=32, DuoGuard commit 44396c3...

initial commit