Posts
All the articles I've posted.
- research
ProVe: Proposer-Verifier Decoding
Speculative decoding with the roles reversed: the large model proposes a block of 16 tokens in one pass, a verifier reads it in one more. 2.5x exact (output equals greedy AR), 2.7x with a 3B verifier.
- dog
Peyto's 1st Year Photos Selection
A photo wall of Peyto's first year, from April 2025 to July 2026, earliest to latest.
- research
MARS: Several Tokens per Forward Pass, Without Giving Up Autoregression
One SFT checkpoint, two modes: full-quality AR at τ=1.0, or 1.8 tokens per forward pass for 1.56–2.16× speedup, with an exact, free fallback to plain AR.
- research
How Much Do Diffusion LLMs Memorize? (vs Autoregressive)
Same size, same memory ceiling, but the diffusion model needs ~10x the training to reach it, and eventually just can't.