Showing 1–3 of 3 · #performance

Oct 28MODELSQuantization without tears: what you lose at 8, 4 and 3 bits6 min ↗Oct 11PAPERSSpeculative decoding: faster answers without changing them6 min ↗Sep 08BENCHMARKSDesigning an LLM latency benchmark that matches what users feel6 min ↗