The ExLlamaV3 project released version 1.4.7, introducing improved CPU mixture-of-experts performance across all tiers and a DRY sampler1. The update also fixes recurrent slot leaks and accelerates quantization for 6-, 7-, and 8-bit trellises, alongside additional bugfixes and quality-of-life improvements.
ExLlamaV3 v1.4.7 Speeds CPU MoE, Adds DRY Sampler
ExLlamaV3 v1.4.7 improves CPU mixture-of-experts performance, adds a DRY sampler, fixes recurrent slot leaks, and accelerates 6/7/8-bit trellis…
