Update llama.py

unsloth/models/llama.py

@@ -1161,7 +1161,7 @@ def CausalLM_fast_forward(fast_forward_inference):
         if not return_dict:
             output = (logits,) + outputs[1:]
             return (loss,) + output if loss is not None else output
-
+        print("***", logits.dtype, logits.shape)
         return CausalLMOutputWithPast(
             loss=loss,
             logits=logits,