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