From 606bc5cd3465d48c66aa573bc1680c9bbe78edd9 Mon Sep 17 00:00:00 2001
From: kijai <40791699+kijai@users.noreply.github.com>
Date: Wed, 4 Mar 2026 15:02:09 +0200
Subject: [PATCH] Accept both torch_dtype and dtype

transformer version dependant
---
 nodes.py | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/nodes.py b/nodes.py
index f6017bb..c054dc0 100644
--- a/nodes.py
+++ b/nodes.py
@@ -206,8 +206,11 @@ class DownloadAndLoadFlorence2Model:
         if version.parse(transformers.__version__) >= version.parse('5.0.0'):
             model, processor = load_model(model_path, attention, dtype, offload_device)
         else:
+            import inspect
             from .modeling_florence2 import Florence2ForConditionalGeneration
-            model = Florence2ForConditionalGeneration.from_pretrained(model_path, attn_implementation=attention, dtype=dtype).to(offload_device)
+            _sig = inspect.signature(Florence2ForConditionalGeneration.from_pretrained)
+            _dtype_kwarg = 'torch_dtype' if 'torch_dtype' in _sig.parameters else 'dtype'
+            model = Florence2ForConditionalGeneration.from_pretrained(model_path, attn_implementation=attention, **{_dtype_kwarg: dtype}).to(offload_device)
             processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)
 
         if lora is not None:
@@ -306,8 +309,11 @@ class Florence2ModelLoader:
         if version.parse(transformers.__version__) >= version.parse('5.0.0'):
             model, processor = load_model(model_path, attention, dtype, offload_device)
         else:
+            import inspect
             from .modeling_florence2 import Florence2ForConditionalGeneration
-            model = Florence2ForConditionalGeneration.from_pretrained(model_path, attn_implementation=attention, dtype=dtype).to(offload_device)
+            _sig = inspect.signature(Florence2ForConditionalGeneration.from_pretrained)
+            _dtype_kwarg = 'torch_dtype' if 'torch_dtype' in _sig.parameters else 'dtype'
+            model = Florence2ForConditionalGeneration.from_pretrained(model_path, attn_implementation=attention, **{_dtype_kwarg: dtype}).to(offload_device)
             processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)
 
         if lora is not None: