Nano-vLLM: A lightweight vLLM implementation built from scratchgithub.com5 points·by platers·anno scorso·0 comments
Synchronizing Pong to music with constrained optimizationvictortao.substack.com321 points·by platers·2 anni fa·37 comments
platers·7 mesi fa·discussthis would pair well with https://gonzoml.substack.com/p/neurips-2025-best-papers-in-c...
platers·11 mesi fa·discussI'm struggling to understand where the gains are coming from. What is the intuition for why DiT training was so inefficient?
platers·11 mesi fa·discusshttps://suno.com/studio-waitlist Just a waitlist so far, but looks like this is the direction suno is going
platers·anno scorso·discussIt sounds very similar to suno v3.5 (including the audio quality) Likely they trained on suno generations.
platers·anno scorso·discussFlash attention natively supports packing multiple variable length sequences into a single call, what is the advantage of jagged flash attention?
platers·2 anni fa·discussIs this wrapping Suno? Not going to make an account with only generic marketing copy and no samples.