Model card: add iPad Pro M4 + iPhone 15 Pro benchmarks

Model card: Pro=distil(fp16) / default=HuBERT positioning

Tier swap: uhm-pro=distil(fp16,Pro), uhm-base=HuBERT(8-bit,default) — uhm-pro-web-fp16.onnx

Tier swap: uhm-pro=distil(fp16,Pro), uhm-base=HuBERT(8-bit,default) — uhm-pro.onnx

Tier swap: uhm-pro=distil(fp16,Pro), uhm-base=HuBERT(8-bit,default) — uhm-base-web-fp16.onnx

Tier swap: uhm-pro=distil(fp16,Pro), uhm-base=HuBERT(8-bit,default) — uhm-base.onnx

Tier swap: uhm-pro=distil(fp16,Pro), uhm-base=HuBERT(8-bit,default) — uhm-pro-8bit.mlpackage.zip

Tier swap: uhm-pro=distil(fp16,Pro), uhm-base=HuBERT(8-bit,default) — uhm-pro.mlpackage.zip

Tier swap: uhm-pro=distil(fp16,Pro), uhm-base=HuBERT(8-bit,default) — uhm-base.mlpackage.zip

Model card: pro v3.x — 8-bit, 137x iPhone 17 Pro / 204x M4 Max

pro v3.x (clearaug enhanced-audio robustness, 8-bit) — uhm-pro-web-fp16.onnx

pro v3.x (clearaug enhanced-audio robustness, 8-bit) — uhm-pro.onnx

pro v3.x (clearaug enhanced-audio robustness, 8-bit) — uhm-pro.mlpackage.zip

Model card: base mlpackage 6-bit ~15 MB

Refresh uhm-base artifacts

Refresh uhm-base artifacts

Refresh uhm-base artifacts

Model card: 6-class indices, 6-bit pro mlpackage, iOS 17 floor

Refresh uhm-pro ONNX

Refresh uhm-pro ONNX

Refresh uhm-pro mlpackage

Refresh uhm-pro artifacts

Refresh uhm-pro artifacts

Refresh uhm-pro artifacts

Model card: v3 6-class metrics + temperature-calibrated artifacts

v3 base retrain — distilled from pro v3, 6-class, T=1.65 calibrated

v3 base retrain — distilled from pro v3, 6-class, T=1.65 calibrated

v3 base retrain — distilled from pro v3, 6-class, T=1.65 calibrated

v3 pro retrain — 6-class (adds and), HARD_NEG_WEIGHT=4, T=2.05 calibrated

v3 pro retrain — 6-class (adds and), HARD_NEG_WEIGHT=4, T=2.05 calibrated

v3 pro retrain — 6-class (adds and), HARD_NEG_WEIGHT=4, T=2.05 calibrated

Upload README.md with huggingface_hub

Upload README.md with huggingface_hub

Update README.md

Upload README.md with huggingface_hub

Add iPhone 17 Pro speed alongside M4 Max

Refresh model card: clean tier table, drop v1/v2 in-flight wording, add Versioning section

Initial release — v2 base + v2 pro under CC BY-NC 4.0

initial commit