Search Results
6/22/2025, 5:57:28 AM
>>105667736
I'll say this again but if they were smart they'd just spin up a 2B or 4B model from scratch using Pixart, similar to what I did, more params, use Sana's Lite Attention and MLP and a 16 channel VAE. An H100 cluster would cruise through it.
I'll say this again but if they were smart they'd just spin up a 2B or 4B model from scratch using Pixart, similar to what I did, more params, use Sana's Lite Attention and MLP and a 16 channel VAE. An H100 cluster would cruise through it.
Page 1