PuzzleMoE: Efficient Compression of Large Mixture-of-Experts Models via Sparse Expert Merging and Bit-packed inferenceRedditJul 07, 2026 03:45Sharesupercomputing-system-ai-lab.github.iosupercomputing-system-ai-lab.github.io