Removed manually sending tensors to devices

HF handles this it is unnecessary and conflicts whne there are multiple gpus

src/deepeval/base_task.py

@@ -53,8 +53,8 @@ class BaseTask(ABC):
             self.tokenizer.pad_token = self.tokenizer.eos_token  # Use EOS token as PAD token
 
         inputs = self.tokenizer(msg, return_tensors="pt", padding=True, truncation=True)
-        input_ids = inputs.input_ids.to(self.model.device)
-        attention_mask = inputs.attention_mask.to(self.model.device)
+        input_ids = inputs.input_ids
+        attention_mask = inputs.attention_mask
 
         if self.model.config.pad_token_id is None:
             self.model.config.pad_token_id = self.tokenizer.eos_token_id
@@ -100,8 +100,8 @@ class BaseTask(ABC):
         formatted_chat = self.tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
         #print(formatted_chat)
         inputs = self.tokenizer(formatted_chat, return_tensors="pt", padding=True, truncation=True)
-        input_ids = inputs.input_ids.to(self.model.device)
-        attention_mask = inputs.attention_mask.to(self.model.device)
+        input_ids = inputs.input_ids
+        attention_mask = inputs.attention_mask
 
         # Generate the sequence of letters starting from 'A'
         letters = [chr(ord('A') + i) for i in range(len(choices))]  # Create option letters A, B, C, D, E, ...
@@ -158,8 +158,8 @@ class BaseTask(ABC):
         )
 
         inputs = self.tokenizer(formatted_chat, return_tensors="pt", padding=True, truncation=True)
-        input_ids = inputs.input_ids.to(self.model.device)
-        attention_mask = inputs.attention_mask.to(self.model.device)
+        input_ids = inputs.input_ids
+        attention_mask = inputs.attention_mask
 
         output = self.model.generate(
             input_ids,