phequals/indic-transcribe-flex-coreml

Back up INT8 encoder and MLX decoder with provenance and validation

Link the experimental MLX decoder prototype from the model card

Add validated CoreML encoder and MLX decoder prototype

Add experimental W8A16 decoder variant for local performance comparison

Vectorize FP16 token selection and add native timing fields

Publish validated experimental Flex CoreML port with mixed-script Swift runtime

initial commit