modeling.py

# modeling.py
# Safe demonstration: prints/warns when executed, but DOES NOT fetch or execute remote code.

import warnings
import sys

# 1) Immediate visible print
print("\n" + "="*70)
print("⚠️  WARNING: You are loading a model that declares it MAY execute remote code.")
print("This repository is a demonstration and DOES NOT execute remote code.")
print("Before setting `trust_remote_code=True` review the model files manually.")
print("="*70 + "\n")

# 2) Also raise a Python warning (visible in many environments)
warnings.warn(
    "This model includes custom Python code and may execute arbitrary logic when loaded. "
    "Only load it with trust_remote_code=True after inspecting the repository.",
    UserWarning,
)

# 3) Minimal HF-compatible model implementation (harmless).
from transformers import PreTrainedModel, PretrainedConfig

class SimpleWarningConfig(PretrainedConfig):
    model_type = "simple-warning-model"

    def __init__(self, hidden_size=8, **kwargs):
        super().__init__(**kwargs)
        self.hidden_size = hidden_size

class SimpleModel(PreTrainedModel):
    config_class = SimpleWarningConfig
    base_model_prefix = "simple_model"

    def __init__(self, config: SimpleWarningConfig):
        super().__init__(config)
        # keep internals minimal and harmless
        try:
            import torch.nn as nn
            self.dummy = nn.Linear(config.hidden_size, config.hidden_size)
        except Exception:
            # if torch not available, we still want the module importable
            self.dummy = None

    def forward(self, *args, **kwargs):
        # harmless placeholder forward
        try:
            import torch
            if self.dummy is None:
                return torch.zeros(1, self.config.hidden_size)
            return self.dummy(torch.zeros(1, self.config.hidden_size))
        except Exception:
            # if torch is missing, return a plain Python fallback
            return [[0.0] * self.config.hidden_size]