Clarify "Load Model" SafeTensors documentation (#3028)

The "Load Model" node description inaccurately implied that SafeTensors
wasn't supported. This was presumably because the documentation in
question was split between two files, and only one was updated.

This commit consolidates the supported file type documentation in one
file, so that this bug is less likely to happen again. This also avoids
making the description tooltip unwieldily verbose, as would have
happened if I had instead edited the description in-place.

Also note in the documentation that SafeTensors should be used for
better security, since it's not really reasonable to expect users to
read the Spandrel GitHub README to find that.

Co-authored-by: Jeremy Rand <[email protected]>

backend/src/nodes/properties/inputs/file_inputs.py

@@ -111,6 +111,9 @@ def PthFileInput(primary_input: bool = False) -> FileInput:
         file_kind="pth",
         filetypes=[".pt", ".pth", ".ckpt", ".safetensors"],
         primary_input=primary_input,
+    ).with_docs(
+        "Supports PyTorch state dict (.pth), TorchScript (.pt), Checkpoint (.ckpt), or SafeTensors (.safetensors) files.",
+        "Use SafeTensors files for better security (the other file types might be able to execute arbitrary code when loaded).",
     )
 
 

backend/src/packages/chaiNNer_pytorch/pytorch/io/load_model.py

@@ -34,10 +34,7 @@ def parse_ckpt_state_dict(checkpoint: dict):
     schema_id="chainner:pytorch:load_model",
     name="Load Model",
     description=[
-        (
-            "Load PyTorch state dict (.pth), TorchScript (.pt), or Checkpoint (.ckpt) files into an"
-            " auto-detected supported model architecture."
-        ),
+        ("Load PyTorch model file into an auto-detected supported model architecture."),
         (
             "- For Super-Resolution, we support most variations of the RRDB"
             " architecture (ESRGAN, Real-ESRGAN, RealSR, BSRGAN, SPSR), Real-ESRGAN's"