CODE - IA - Python - How to convert PyTorch `.pth` tensors to the SafeTensors format

import torch
from safetensors.torch import save_file

# Load your PyTorch model or state dict
pytorch_model = torch.load("your_model.pth")

# If it's a full model, you might need to get the state dict
if hasattr(pytorch_model, "state_dict"):
    state_dict = pytorch_model.state_dict()
else:
    state_dict = pytorch_model

# Save as safetensors format
save_file(state_dict, "converted_model.safetensors")


Explanation:
1. **Import necessary libraries**:
   ```python
   import torch
   from safetensors.torch import save_file
   ```
   - `torch` is the PyTorch library
   - `save_file` is a function from the safetensors library specifically designed for saving PyTorch tensors

2. **Load the original PyTorch model**:
   ```python
   pytorch_model = torch.load("your_model.pth")
   ```
   This loads your model or state dictionary from a .pth file into memory.

3. **Extract the state dictionary**:
   ```python
   if hasattr(pytorch_model, "state_dict"):
       state_dict = pytorch_model.state_dict()
   else:
       state_dict = pytorch_model
   ```
   This checks if what you loaded is a full model object (which has a state_dict() method) or already a state dictionary. The state dictionary is a Python dictionary mapping parameter names to their tensor values.

4. **Save in SafeTensors format**:
   ```python
   save_file(state_dict, "converted_model.safetensors")
   ```
   This converts your PyTorch state dictionary to the SafeTensors format and saves it to a file.

SafeTensors is designed to be more secure and efficient than PyTorch's native format, providing protection against certain security vulnerabilities while potentially offering faster loading times.