Tag
A community proposal in r/LocalLLaMA urging members to stop fixating on tok/s speed metrics and instead create private quality benchmarks, especially to evaluate model quality across quantisation levels, and to share results with each other.