v0.1.9a3

What's Changed

• [model, ckpt] feat: runtime checkpoint tensor converter for transformers v4 MoE by @Luosuu in #696
• [model] fix: qwen3_vl vision dummy_forward sp shape mismatch by @TimYangst in #697
• [misc] feat: add ep_size property to ParallelState by @deerlu in #706
• [model, ops, trainer] refactor: unify MoE dispatch through OpSlot by @TimYangst in #705
• [ops] fix: use reshape for non-contiguous chunks in chunk_loss by @TimYangst in #708
• [parallel, ci] feat: support loading large tensor for fsdp2 by @JorgenWan in #688
• [ops, trainer] feat: chunked per-token log-probs for PPO-style RL by @Luosuu in #711
• [release] chore: release v0.1.9a3 by @TimYangst in #709

Full Changelog: v0.1.9a2...v0.1.9a3