Show HN: 8-bit Qwen3.8-27B decodes 1.7x faster than BF16, slower at 16K context

(github.com)

1 points | by gltanaka 6 hours ago ago

No comments yet.