Nano-vLLM: A lightweight vLLM implementation built from scratchgithub.com5 points·by platers·ano passado·0 comments
platers·há 7 meses·discussthis would pair well with https://gonzoml.substack.com/p/neurips-2025-best-papers-in-c...
platers·há 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·há 11 meses·discusshttps://suno.com/studio-waitlist Just a waitlist so far, but looks like this is the direction suno is going
platers·ano passado·discussIt sounds very similar to suno v3.5 (including the audio quality) Likely they trained on suno generations.
platers·ano passado·discussFlash attention natively supports packing multiple variable length sequences into a single call, what is the advantage of jagged flash attention?