Qwen3-30b learn to init

[gagika]

Description

Qwen3 MoE: Learn-to-Init distillation

Adds LTI soft distillation for Qwen3-30B-A3B-base MoE → custom student. Wires the LTI hook into Qwen3MoeDecoderLayerToLinen, adds warm-start init for the A/B bridges, and ships a recipe + setup doc.

Tests

• End-to-end TPU v7x-8 run: ~12K steps, KL converges to ~0.41

gs://agagik-us/distillation/qwen3-30b-lti/qwen3-30b-lti-rope1e7-20260518-181330/checkpoints/

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

[codecov]

Codecov Report

❌ Patch coverage is 32.25806% with 21 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/maxtext/layers/learn_to_init_layer.py	25.00%	21 Missing ⚠️

📢 Thoughts on this report? Let us know!