4 points | by danielhanchen 14 hours ago
1 comments
Hey HN! Just sharing some work we did to make gpt-oss finetuning use O(N) and not O(N^2) VRAM via Flex Attention + some bug fixes :)
Hey HN! Just sharing some work we did to make gpt-oss finetuning use O(N) and not O(N^2) VRAM via Flex Attention + some bug fixes :)