Support new model Qwen/Qwen3.6-35B-A3B

[lvliang-intel]

Description

Support new model Qwen/Qwen3.6-35B-A3B.
Qwen/Qwen3.6-35B-A3B still use the same model type with Qwen/Qwen3.5-35B-A3B.
This PR adds compatibility for Qwen/Qwen3.6-35B-A3B in the AutoRound HF conversion path.

Type of Change

• Bug fix
• New feature
• Documentation update
• Performance improvement
• Code refactoring
• Other (please specify):

Related Issues

Fixes or relates to #

Checklist Before Submitting

• My code has been tested locally.
• Documentation has been updated as needed.
• New or updated tests are included where applicable.

[Copilot]

Pull request overview

Adds compatibility for the new HuggingFace model Qwen/Qwen3.6-35B-A3B by improving AutoRound’s HF conversion logic when a composite checkpoint is loaded as a text-only submodel and module path prefixes differ.

Changes:

• Added _remap_paths_for_text_model() to remap quantization block paths using Transformers checkpoint conversion mappings.
• Updated get_layer_config() to retry layer discovery with remapped paths when the initial block-path match finds zero layers.

[wenhuach21]

Shall we add version control if it's related to Transformer bugs, so we can handle them cleanly if they’re fixed in future releases

[lvliang-intel]

Shall we add version control if it's related to Transformer bugs, so we can handle them cleanly if they’re fixed in future releases

This issue does not occur when using Qwen3_5MoeForConditionalGeneration.from_pretrained; it only appears with AutoModelForCausalLM.from_pretrained. So it's not a real issue? If users want to use AutoModelForCausalLM.from_pretrained to load Qwen/Qwen3.6-35B-A3B models, they need to this PR and transformers issue fix huggingface/transformers#45494.

[lvliang-intel]

/azp run Unit-Test-CUDA-AutoRound

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).