perf(qwen3-next): nested scan over hybrid attention with per-layer remat - #4964

perf(qwen3-next): nested scan over hybrid attention with per-layer remat#4964

copybara-service[bot] merged 11 commits into

mainAI-Hypercomputer/maxtext:mainfrom

chengnuojin-bharatgen-scanAI-Hypercomputer/maxtext:chengnuojin-bharatgen-scanCopy head branch name to clipboard