== models--google--gemma-4-E2B-it
  bytes read per decode token: 4.597 GB
  bound @ 277.0 GB/s measured stream:   60.3 tok/s
== models--google--gemma-4-E2B-it-qat-w4a16-ct
  bytes read per decode token: 1.862 GB
  bound @ 277.0 GB/s measured stream:  148.8 tok/s