tmxkzm1925-max/MoE-Direct

Byte-preserving MoE expert streaming from NVMe — serve MoE models larger than RAM on a consumer Windows PC, with verified token-identical outputs

1 best points · 0 comments · 1 Hacker News post · 11 GitHub stars · View on GitHub

Hacker News discussions

• Show HN: Moe-Direct – MoE Models far larger than your RAM, on a consumer desktop (1 points, 0 comments, by tmxkzm1925-max, 2026-08-29)