Add streaming float and int8 graphs

Model card: separate trained parameters from fixed values in the weights file

Add the full distillation checkpoints (step 26,000 and 30,000)

Model card: base model, datasets, languages, tags and evaluation results in the metadata

Add PyTorch weights (safetensors) and model definition

Licence matches the teacher's; card and config

Model card: causal, 100 ms delay, streaming context

WakeHuBERT tiny extractor

initial commit