What does this PR do?

Fixes #14195

_best_guess_weight_name() currently checks local_files_only and HF_HUB_OFFLINE before determining whether the supplied path is a local file or directory. As a result, local LoRA weights without an explicit weight_name are rejected in offline mode before Diffusers inspects the local filesystem.

This PR moves the offline-mode guard after the local file and directory checks. With this change:

• a direct local LoRA file works in offline mode;
• a local directory can auto-discover its LoRA weight file with either HF_HUB_OFFLINE=1 or local_files_only=True;
• a remote repository ID without weight_name still raises the existing offline-mode error;
• model_info() is not called in offline mode.

Checking os.path.isfile() and os.path.isdir() only accesses the local filesystem and cannot trigger a Hub request, so this preserves the network restrictions of offline mode. There are no public API changes.

Coordination: #14195 and the related discussion in #14198.

Tests

The regression tests cover local files, local directories, both offline controls, remote repository IDs without weight_name, the absence of model_info() calls, and the existing behavior for zero or multiple matching files.

$ .venv/bin/python -m pytest -q tests/lora/test_lora_loader_utils.py
......                                                                   [100%]
6 passed in 0.05s

$ PATH=".venv/bin:$PATH" make style
All checks passed!
1999 files left unchanged

$ PATH=".venv/bin:$PATH" make fix-copies
Passed

$ PATH=".venv/bin:$PATH" make quality
All checks passed!
1999 files already formatted

AI assistance

This PR received substantial assistance from OpenAI Codex. I identified and reproduced the original problem, defined the expected behavior and regression cases, and made the final contribution decisions. Codex was used to inspect the repository instructions and related history, implement the minimal fix, add regression tests, run the relevant checks, perform the final self-review, and help draft the PR description. The final diff and test results were presented to me before submission.

Final self-review

• Blocking issues: None.
• Non-blocking issues: None.
• Dead code: None introduced; all new test helpers and constants are used.
• Documentation impact: No documentation update is needed. This is a focused bug fix with no public API or default changes, and remote repository behavior remains unchanged.
• Verdict: READY.
• Fix before submitting: Nothing.
• Leave for the actual review: Nothing.

Before submitting

• Did you use an AI agent (Claude Code, Codex, Cursor, etc.) to help with this PR? If so:
  • Did you read the Coding with AI agents guide?
  • Did you run the self-review skill on the diff?
  • Did you share the final self-review notes in the PR description or a comment?
• Did you read the contributor guideline?
• Did you read our philosophy doc? (important for complex PRs)
• Was this discussed/approved via a GitHub issue or the forum? Discussed in Offline mode rejects local LoRA files and directories without weight_name #14195 and fix: allow local LoRA paths in offline mode #14198.
• Did you make sure to update the documentation with your changes? No update is needed for this focused bug fix; the public API is unchanged.
• Did you write any new necessary tests?
• Are you the author (or part of the team) of the model/pipeline? Not applicable; this PR does not add a model or pipeline.

Who can review?

@sayakpaul