PR #44916: fix(deberta-v2): move "Copied from" comments above @torch.jit.script for Python 3.13 compat

• Repository: huggingface/transformers
• Author: s-zx
• State: closed | merged: False
• Link: https://github.com/huggingface/transformers/pull/44916

Description (problem / solution / changelog)

Summary

Importing DebertaV2Model (or anything that depends on it, e.g. gliner) raises IndentationError on Python 3.13 because torch.jit.script calls inspect.getsource(), dedents the snippet, and passes it to ast.parse(). Python 3.13's stricter parser rejects a comment placed between a decorator and the def statement.

Root Cause

Three functions in modeling_deberta_v2.py had a # Copied from ... comment between @torch.jit.script and def:

@torch.jit.script
# Copied from transformers.models.deberta.modeling_deberta.c2p_dynamic_expand
def c2p_dynamic_expand(...):
    ...

Fix

Move each # Copied from comment to the line above its @torch.jit.script decorator:

+# Copied from transformers.models.deberta.modeling_deberta.c2p_dynamic_expand
 @torch.jit.script
-# Copied from transformers.models.deberta.modeling_deberta.c2p_dynamic_expand
 def c2p_dynamic_expand(...):

Applied to all three affected functions (c2p_dynamic_expand, p2c_dynamic_expand, pos_dynamic_expand).

Fixes #44855

Changed files

• src/transformers/models/deberta_v2/modeling_deberta_v2.py (modified, +3/-3)
• src/transformers/models/gpt_neox/configuration_gpt_neox.py (modified, +5/-1)

PR #44917: fix(gpt-neox): preserve rotary_pct across save/load cycle

• Repository: huggingface/transformers
• Author: s-zx
• State: closed | merged: False
• Link: https://github.com/huggingface/transformers/pull/44917

Description (problem / solution / changelog)

Summary

GPTNeoXConfig.convert_rope_params_to_dict unconditionally overwrote rope_parameters["partial_rotary_factor"] with the default 0.25 when rotary_pct was absent from kwargs.

On every from_pretrained call, rotary_pct is not present as a top-level key in config.json (it is stored inside rope_parameters), so the value was silently reset to 0.25.

Fix

Replace the unconditional assignment with a conditional + setdefault:

-self.rope_parameters["partial_rotary_factor"] = kwargs.pop("rotary_pct", 0.25)
+rotary_pct = kwargs.pop("rotary_pct", None)
+if rotary_pct is not None:
+    self.rope_parameters["partial_rotary_factor"] = rotary_pct
+else:
+    self.rope_parameters.setdefault("partial_rotary_factor", 0.25)

When rotary_pct is absent (the reload path), setdefault only fills in 0.25 if partial_rotary_factor is not already present, so an existing value is preserved.

Fixes #44913

Changed files

• src/transformers/models/gpt_neox/configuration_gpt_neox.py (modified, +5/-1)

PR #44986: fix: remove Copied from comments between @torch.jit.script and def for Python 3.13 compat

• Repository: huggingface/transformers
• Author: Krishnachaitanyakc
• State: closed | merged: True
• Link: https://github.com/huggingface/transformers/pull/44986

Description (problem / solution / changelog)

Summary

Fixes #44855

On Python 3.13, placing a # Copied from comment between @torch.jit.script and the function definition causes an IndentationError. This happens because torch.jit.script calls inspect.getsource() followed by ast.parse(), and Python 3.13's stricter parser fails to associate the function body with the def when a comment intervenes between the decorator and the function signature.

As suggested by @Rocketknight1, the fix is to remove the # Copied from comments entirely from these locations. The three affected functions (c2p_dynamic_expand, p2c_dynamic_expand, pos_dynamic_expand) are identical to their source in modeling_deberta.py and will remain in sync without the copy markers.

Changes

• modeling_deberta_v2.py: Removed 3 # Copied from comments between @torch.jit.script decorators and def statements
• modeling_sew_d.py: Removed the same 3 # Copied from comments (same bug pattern)

Coordination

• Issue discussion and maintainer approval: https://github.com/huggingface/transformers/issues/44855#issuecomment-4122734140
• Previous PRs #44856 and #44916 were closed because they moved the comments instead of removing them, which is not what was requested
• No existing open PRs address this fix

Testing

• Verified DebertaV2Model imports successfully after the change
• Verified sew_d module imports successfully after the change
• Ruff linting and formatting checks pass

AI assistance disclosure

This PR was prepared with AI assistance (Claude Code). The change was reviewed line-by-line by the submitter before opening.

Changed files

• src/transformers/models/deberta_v2/modeling_deberta_v2.py (modified, +0/-3)
• src/transformers/models/sew_d/modeling_sew_d.py (modified, +0/-3)