diff --git a/src/llmtuner/data/template.py b/src/llmtuner/data/template.py
index 631c79c1..0b2ca0e6 100644
--- a/src/llmtuner/data/template.py
+++ b/src/llmtuner/data/template.py
@@ -859,7 +859,7 @@ _register_template(
 _register_template(
     name="yivl",
     format_user=StringFormatter(slots=["### Human: {{content}}\n### Assistant:"]),
-    format_assistant=StringFormatter(slots=[" {{content}}"]),
+    format_assistant=StringFormatter(slots=[" {{content}}\n"]),
     stop_words=["###"],
     default_system=(
         "This is a chat between an inquisitive human and an AI assistant. "
diff --git a/src/llmtuner/model/utils/visual.py b/src/llmtuner/model/utils/visual.py
index 79a6570e..8553cf86 100644
--- a/src/llmtuner/model/utils/visual.py
+++ b/src/llmtuner/model/utils/visual.py
@@ -29,7 +29,8 @@ def autocast_projector_dtype(
     ) -> "torch.Tensor":
         return output.to(model_args.compute_dtype)
 
-    if hasattr(model, mm_projector_name) and getattr(model.config, "quantization_method", None):
+    if hasattr(model, mm_projector_name) and (getattr(model.config, "quantization_method", None)
+                                              or "Yi" in getattr(model.config.text_config, "_name_or_path", None)):
         logger.info("Casting multimodal projector outputs in {}.".format(model_args.compute_dtype))
         mm_projector: "torch.nn.Module" = getattr(model, mm_projector_name)
         mm_projector.register_forward_hook(_mm_projector_forward_post_hook)