Specter: Fast MoE Inference via Quantization-Enabled Multi-Step Speculative Offloading Paper • 2511.14102 • Published 3 days ago • 21
YuE2: Unifying Symbolic and Audio Music Generation at Frontier Quality Paper • 2609.33757 • Published 26 days ago • 206
HOBBIT: A Mixed Precision Expert Offloading System for Fast MoE Inference Paper • 2411.01433 • Published Nov 3, 2024 • 2