Nano-vLLM: A lightweight vLLM implementation built from scratchgithub.com5 points·by platers·el año pasado·0 comments
platers·hace 7 meses·discussthis would pair well with https://gonzoml.substack.com/p/neurips-2025-best-papers-in-c...
platers·hace 11 meses·discussI'm struggling to understand where the gains are coming from. What is the intuition for why DiT training was so inefficient?
platers·hace 11 meses·discusshttps://suno.com/studio-waitlist Just a waitlist so far, but looks like this is the direction suno is going
platers·el año pasado·discussIt sounds very similar to suno v3.5 (including the audio quality) Likely they trained on suno generations.
platers·el año pasado·discussFlash attention natively supports packing multiple variable length sequences into a single call, what is the advantage of jagged flash attention?