instruct model setup

unsloth_compiled_cache/UnslothGRPOTrainer.py

@@ -1,7 +1,7 @@
 """
-2025.8.4
-2025.8.5
-4.55.1
+2025.8.9
+2025.8.10
+4.55.4
 0.21.0
 __UNSLOTH_VERSIONING__
 """
@@ -727,6 +727,7 @@ class UnslothGRPOConfig(GRPOConfig):
         default = -1,
         metadata = {'help': 'Chunk size to reduce memory usage. -1 is most efficient.'},
     )
+    
     def __init__(
         self,
         output_dir = None,
@@ -904,6 +905,7 @@ class UnslothGRPOConfig(GRPOConfig):
         wandb_log_unique_prompts = False,
         vllm_sampling_params = None,
         unsloth_num_chunks = -1,
+        
         **kwargs,
     ):
         if learning_rate < 1e-7: raise FloatingPointError(f'Unsloth: Your learning rate of `{learning_rate}` is too small and less than 1e-7! Consider increasing it, otherwise gradient updates will be close to 0!')
@@ -1116,6 +1118,7 @@ class UnslothGRPOConfig(GRPOConfig):
             wandb_log_unique_prompts = wandb_log_unique_prompts,**kwargs)
         self.vllm_sampling_params = vllm_sampling_params
         self.unsloth_num_chunks = unsloth_num_chunks
+        
 pass
 
 class _UnslothGRPOTrainer(Trainer):
@@ -2814,7 +2817,7 @@ class UnslothGRPOTrainer(_UnslothGRPOTrainer):
             print('Unsloth: Switching to float32 training since model cannot work with float16')
             force_float32 = True
         mixed_precision_dtype = os.environ.get('UNSLOTH_MIXED_PRECISION', 'float32')
-        dtype = getattr(model.config, 'torch_dtype', None)
+        dtype = getattr(model.config, 'dtype', None) or getattr(model.config, 'torch_dtype', None)
         if dtype is None: dtype = model.get_input_embeddings().dtype
         from unsloth_zoo.utils import _get_dtype
         dtype = _get_dtype(dtype)