Compressed models are NOT miniature versions of large models

Rohit Raj Rai, Rishant Pal, Amit Awekar

Large neural models are often compressed before deployment. Model compression is necessary for many practical reasons, such as inference latency, memory footprint, and energy consu…