Zarya: A Hybrid Autoregressive--Masked Diffusion Language Model with Flexible Training and Dual-Mode Inference Paper • 2609.19868 • Published 1 day ago
How Lossless Is Lossless Speculative Decoding? The Role of Numerical Precision in Orthrus Paper • 2609.15504 • Published 5 days ago • 31
How Lossless Is Lossless Speculative Decoding? The Role of Numerical Precision in Orthrus Paper • 2609.15504 • Published 5 days ago • 31