software-mansion/react-native-executorch-efficientnet-v2-s

int8: keep squeeze-excitation fp32, Imagenette calibration (ImageNetV2 top-1 72.1% -> 75.0%)

QNN: re-export a16w8 with MinMaxObserver activations

Add QNN a16w8 exports, one per Hexagon version (v69-v81)

Add Vulkan fp16 exports

Apply the model card standard

Apply the model card standard

Stop restating quantized, and drop the unused default

Correct the published metadata

Core ML: drop the fp32 variant, fp16 is equivalent in task terms

NOTES: correct the availability-history claim

Remove MLX artifacts: 7.5-19.6x slower than Core ML on device

Add per-backend config

Add mlx export (ExecuTorch 1.4.1, fork toolchain)

Add stub root config.json for HF download counter

Add spec-compliant config.json files

Add spec-compliant config.json files

Remove old-layout metadata orphaned by MODEL_SPEC.md restructure

Update README

Update models. Add quantized models.

Update .gitattributes

Delete xnnpack/fsmn-vad_xnnpack.pte

Upload fsmn-vad_xnnpack.pte

Update README.md

Update README.md

Update README.md

Create config.json

Update README.md

Upload models

initial commit