modular_model_converter can not handle objects import via try - except

[HIT-cwh]

System Info

transformers 4.48.0.dev0 d8c1db2

Who can help?

@ArthurZucker

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

How to reproduce?

1. Clone the Transformers repository and check out the specified commit:

git clone [email protected]:huggingface/transformers.git && cd transformers && git checkout d8c1db2f568d4bcc254bc046036acf0d6bba8373

2. Create a new folder named xxx_model in src/transformers/models/

3. Inside this folder, create a new Python file called modular_xxx.py with the following content:

import torch
import torch.nn as nn
try:
    import torch.nn.functional as F
except:
    pass

from ..llama.modeling_llama import (
    LlamaMLP,
)

class Model(nn.Module):
    def forward(self, x, w):
        return F.linear(x, w)

4. Run the following command to execute the model converter:
   python utils/modular_model_converter.py --files_to_parse src/transformers/models/xxx_model/modular_xxx.py

This will generate the modeling file at: src/transformers/models/xxx_model/modeling_xxx.py.

Expected behavior

Expected vs Actual Contents in src/transformers/models/xxx_model/modeling_xxx.py

The expected contents in src/transformers/models/xxx_model/modeling_xxx.py is :

#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
#           This file was automatically generated from src/transformers/models/xxx_model/modular_xxx.py.
#               Do NOT edit this file manually as any edits will be overwritten by the generation of
#             the file from the modular. If any change should be done, please apply the change to the
#                          modular_xxx.py file directly. One of our CI enforces this.
#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
import torch.nn as nn

try:
    import torch.nn.functional as F
except:
    pass

class Model(nn.Module):
    def forward(self, x, w):
        return F.linear(x, w)

However, the actual content generated in src/transformers/models/xxx_model/modeling_xxx.py is :

#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
#           This file was automatically generated from src/transformers/models/xxx_model/modular_xxx.py.
#               Do NOT edit this file manually as any edits will be overwritten by the generation of
#             the file from the modular. If any change should be done, please apply the change to the
#                          modular_xxx.py file directly. One of our CI enforces this.
#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
import torch.nn as nn


class Model(nn.Module):
    def forward(self, x, w):
        return F.linear(x, w)

Issue

The lines try: import torch.nn.functional as F except: pass are missing in the actual content, even though it exists in the original modular file.