fix gguf bug for qwen3.5 moe and some small models

[n1ck-guo]

Description

Please briefly describe your main changes, the motivation.

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

This PR targets GGUF export robustness and inference correctness (notably for Qwen3.5-series models) by aligning tensor handling with llama.cpp expectations and preventing failures in edge cases during GGUF conversion/packing.

Changes:

• Guard GGUF tensor preparation against empty tensor_map.mapping (e.g., block_count=0 models) to avoid crashes.
• Expand the “always-F32” tensor-type allowlist during GGUF export to include additional llama.cpp tensor identifiers.
• Adjust RTN layer quantization flow for immediate GGUF packing so the GGUF-specific path is taken regardless of disable_opt_rtn.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
auto_round/export/export_to_gguf/convert.py	Prevents failures when tensor_map.mapping is empty; updates GGUF tensor-type handling for F32 selection.
auto_round/compressors/base.py	Changes RTN quantization branching for immediate GGUF packing, affecting how disable_opt_rtn influences behavior.

[wenhuach21]

Great！been waiting for this pr for a long time

[wenhuach21]

python3 -m auto_round Qwen/Qwen3-1.7B    --enable_alg_ext --iters 1 --nsamples 1 --format gguf:q2_k_s

File "/home/wenhuach/auto-round-main/auto_round/main.py", line 827, in
run()
~~~^^
File "/home/wenhuach/auto-round-main/auto_round/main.py", line 811, in run
start()
~~~~~^^
File "/home/wenhuach/auto-round-main/auto_round/main.py", line 533, in start
tune(args)
~~~~^^^^^^
File "/home/wenhuach/auto-round-main/auto_round/main.py", line 750, in tune
model, folders = autoround.quantize_and_save(export_dir, format=args.format) # pylint: disable=E1101
~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/wenhuach/auto-round-main/auto_round/compressors/base.py", line 1004, in quantize_and_save
model, folders = self.save_quantized(
~~~~~~~~~~~~~~~~~~~^
output_dir, format=self.formats, inplace=inplace, return_folders=True, **kwargs
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/home/wenhuach/auto-round-main/auto_round/compressors/base.py", line 3375, in save_quantized
compressed_model = format.save_quantized(
save_folder,
...<6 lines>...
**kwargs,
)
File "/home/wenhuach/auto-round-main/auto_round/formats.py", line 829, in save_quantized
return save_quantized_as_gguf(
output_dir=output_dir,
...<6 lines>...
**kwargs,
)
File "/home/wenhuach/miniforge3/envs/autoround/lib/python3.13/site-packages/torch/utils/_contextlib.py", line 124, in decorate_context
return func(*args, **kwargs)
File "/home/wenhuach/auto-round-main/auto_round/export/export_to_gguf/export.py", line 266, in save_quantized_as_gguf
gguf_model.write()
~~~~~~~~~~~~~~~~^^
File "/home/wenhuach/auto-round-main/auto_round/export/export_to_gguf/convert_hf_to_gguf.py", line 750, in write
self.prepare_tensors()
~~~~~~~~~~~~~~~~~~~~^^
File "/home/wenhuach/auto-round-main/auto_round/export/export_to_gguf/convert.py", line 653, in prepare_tensors
data, data_qtype = _quant_data(
~~~~~~~~~~~^
cls, data_torch, data_qtype, name, modify_name, new_name, bid, device=device
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/home/wenhuach/auto-round-main/auto_round/export/export_to_gguf/convert.py", line 301, in _quant_data
data = ggml_quant(data_torch, data_qtype.name.lower(), device=device, **kwargs)
File "/home/wenhuach/auto-round-main/auto_round/export/export_to_gguf/packing.py", line 105, in ggml_quant
new_data = ggml_quant_core(quant_func, blocks, scale, zp, wmin, d_scale, d_wmin, imatrix, original)
File "/home/wenhuach/auto-round-main/auto_round/export/export_to_gguf/packing.py", line 35, in ggml_quant_core
new_data = quant_func(
blocks,
...<6 lines>...
original=original,
)
File "/home/wenhuach/auto-round-main/auto_round/export/export_to_gguf/packing.py", line 615, in q2_k_quant_block
mins = wmin.reshape((-1, QK_K // 16))
^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'reshape'

[n1ck-guo]

/azp run Unit-Test-CUDA-AutoRound

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).