Qwen3.5 vision layers ckpt conversion and decode

[hengtaoguo]

Description

This PR supports Qwen3.5-35B model's vision tower checkpoint conversion and enables decode. The JAX layers have been implemented in #3962 , mostly inherited from Qwen3-Omni. The deltas in this PR:

1. Added vision configs for qwen3.5-25b-a3b.yml.
2. Added vision tower parameter mapping for Qwen3.5 in param_mapping.py for weights conversion.
3. Register qwen3.5-25b-a3b throughout the multimodal workflow.
4. Reuse Qwen3-Omni preprocessing utils since they are almost identical.
5. Important Updates: Introduce QwenTokens, since Qwen3.5 changed the token_ids for some special multimodal tokens. QwenTokens will default to Qwen3-Omni so it doesn't impact existing flows.

Tests

Checkpoint conversion:

# Command:
export JAX_PLATFORMS=cpu
python -m maxtext.checkpoint_conversion.to_maxtext src/maxtext/configs/base.yml model_name=qwen3.5-35b-a3b base_output_directory=gs://hengtaoguo-maxtext-logs/checkpoints/qwen3.5-35b-a3b/unscanned/2026-06-01-15-15 use_multimodal=true scan_layers=false hf_access_token=<your_token>

Multimodal decode:

# Command
python -m maxtext.inference.decode maxtext/configs/base.yml model_name=qwen3.5-35b-a3b tokenizer_path=Qwen/Qwen3.5-35B-A3B tokenizer_type=huggingface load_parameters_path=gs://hengtaoguo-maxtext-logs/checkpoints/qwen3.5-35b-a3b/unscanned/2026-06-01-15-15/0/items per_device_batch_size=1 run_name=ht_test_decode_full steps=1 async_checkpointing=false scan_layers=false use_multimodal=true prompt=\'Describe\ this\ image\ in\ one\ sentence.\' image_path=\'tests/assets/test_image.jpg\' max_prefill_predict_length=512 max_target_length=542 attention=\'dot_product\' hf_access_token=<your_token> add_bos=False

# Result
Input `<|im_start|>user
<|vision_start|><|image_pad|><|vision_end|>Describe this image in one sentence.<|im_end|>
<|im_start|>assistant
` -> `<think>

</think>

This vibrant cityscape showcases Seattle’s iconic skyline under a bright blue sky, with the Space Needle standing tall among modern skyscrapers,`

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 3.57143% with 135 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...xtext/checkpoint_conversion/utils/param_mapping.py	0.00%	92 Missing ⚠️
src/maxtext/multimodal/processor_qwen3_omni.py	12.50%	28 Missing ⚠️
src/maxtext/multimodal/processor.py	6.66%	14 Missing ⚠️
src/maxtext/layers/encoders.py	0.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!