v0.1.9a5

What's Changed

• [ci, model] test: add bitwise logits-equal tests for v5 models by @TimYangst in #722
• [ci] refactor: split GPU unit tests into v4/v5 parallel jobs by @TimYangst in #725
• [BREAKING][ops, model] feat: GPU-optimal ops defaults + strict NPU validation by @TimYangst in #716
• [model] fix: fix flops count by @FoolPlayer in #730
• [model] fix: preserve FSDP2 pre-backward hooks for log-prob outputs in qwen3_5_moe and other models by @deerlu in #731
• [ci] refactor: split GPU e2e tests into v4/v5 parallel jobs by @TimYangst in #733
• [docker] fix: datasets version fix by @phdddd in #691
• [release] chore: release v0.1.9a5 by @deerlu in #736

Full Changelog: v0.1.9a4...v0.1.9a5