Migrate Qwen3-Next and Qwen3.5 models to HybridModel

[Phlip79]

Summary

• migrate Qwen3-Next to HybridModel
• migrate the reusable dense and MoE Qwen3.5 text-decoder architectures to HybridModel
• share Gated DeltaNet/full-attention schedule translation and logical-to-physical layer mappings
• update focused bridge tests for both model families

Models modified

Qwen3-Next checkpoint using Qwen3NextForCausalLM:

• Qwen/Qwen3-Next-80B-A3B-Instruct

Qwen3.5 text-decoder architectures:

• qwen3_5_text using Qwen3_5ForCausalLM
• qwen3_5_moe_text using Qwen3_5MoeForCausalLM

These are reusable language-decoder architectures, not separate public Qwen3.5 checkpoint repositories. They appear as the nested text_config inside the official Qwen/Qwen3.5-* multimodal checkpoints. This PR migrates those decoder bridges and provides language-layer mapping helpers reused by the full multimodal bridges.

The official Qwen/Qwen3.5-* checkpoint repositories, their top-level conditional-generation wrappers, vision models, and end-to-end conversion paths are modified in #4837.

Stack

This is PR 2 of 4. Merge the stack in order:

1. Migrate Qwen3 dense and MoE models to HybridModel #4747 - Qwen3 dense and MoE
2. Migrate Qwen3-Next and Qwen3.5 models to HybridModel #4836 - Qwen3-Next and Qwen3.5 text
3. Migrate Qwen VL models to HybridModel #4837 - Qwen3-VL and Qwen3.5-VL
4. Migrate Qwen3-ASR and Qwen3-Omni to HybridModel #4838 - Qwen3-ASR and Qwen3-Omni

Validation

• uv run --no-project --with pre-commit pre-commit run --all-files
• The complete four-PR stack is tree-identical to the original Migrate Qwen3 dense and MoE models to HybridModel #4747 implementation.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.