fixed off by one error on masking (#1193)

Signed-off-by: Dillon Laird <dillonalaird@gmail.com>

intel_extension_for_transformers/neural_chat/examples/finetuning/multi_modal/llava_utils.py

@@ -335,10 +335,10 @@ def preprocess_v1(
 
             if has_image:
                 round_len = len(tokenizer_image_token(rou, tokenizer))
-                instruction_len = len(tokenizer_image_token(parts[0], tokenizer)) - 1
+                instruction_len = len(tokenizer_image_token(parts[0], tokenizer)) - 2
             else:
                 round_len = len(tokenizer(rou).input_ids)
-                instruction_len = len(tokenizer(parts[0]).input_ids) - 1
+                instruction_len = len(tokenizer(parts[0]).input_ids) - 2
 
             target[cur_len : cur_len + instruction_len] = IGNORE_INDEX