Add OVIS 2.5 Support

[Qubitium]

Fix #1971

[Qubitium]

@avtc I have added barriers so the main loop will wait for all threads to complete work before any forward and forward (replay). This may reduce your chance of OOM. Another toggle you can try is changing the cuda:per value in module_looper.py from 4 to 1 to only allow one work task per gpu:index device.

[avtc]

@Qubitium

Traceback (most recent call last):
  File "/home/ubuntu/Documents/Quantize/quantize-glm4.5-Air-gptqmodel-moe-prune-smart-4.py", line 462, in <module>
    from gptqmodel import GPTQModel, QuantizeConfig
  File "/home/ubuntu/git/avtc/GPTQModel/gptqmodel/__init__.py", line 11, in <module>
    from .models import GPTQModel, get_best_device
  File "/home/ubuntu/git/avtc/GPTQModel/gptqmodel/models/__init__.py", line 7, in <module>
    from .auto import MODEL_MAP, GPTQModel
  File "/home/ubuntu/git/avtc/GPTQModel/gptqmodel/models/auto.py", line 60, in <module>
    from .definitions.apertus import ApertusQModel  # noqa: E402
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/ubuntu/git/avtc/GPTQModel/gptqmodel/models/definitions/__init__.py", line 47, in <module>
    from .ovis2_5 import Ovis2_5QModel
ModuleNotFoundError: No module named 'gptqmodel.models.definitions.ovis2_5'

[Qubitium]

@avtc oof. comment that line out. Forgot to git add the new file.

[avtc]

Will check tomorrow