Zarya: A Hybrid Autoregressive--Masked Diffusion Language Model with Flexible Training and Dual-Mode Inference Paper • 2609.19868 • Published 3 days ago
How Lossless Is Lossless Speculative Decoding? The Role of Numerical Precision in Orthrus Paper • 2609.15504 • Published 6 days ago • 33
How Lossless Is Lossless Speculative Decoding? The Role of Numerical Precision in Orthrus Paper • 2609.15504 • Published 6 days ago • 33