diff --git a/JC2.py b/JC2.py
index 269cb1f..10aedca 100644
--- a/JC2.py
+++ b/JC2.py
@@ -111,6 +111,8 @@ def load_models(model_path, dtype, device="cuda", max_memory=None):
 
     try:
         if dtype == "nf4":
+            assert torch.backends.mps.is_available() == False, "NF4 is not currently supported on MPS/Apple Silicon"
+            
             from transformers import BitsAndBytesConfig
             nf4_config = BitsAndBytesConfig(
                 load_in_4bit=True, 
@@ -138,8 +140,8 @@ def load_models(model_path, dtype, device="cuda", max_memory=None):
             text_model = AutoModelForCausalLM.from_pretrained(
                 model_path, 
                 quantization_config=nf4_config,
-                device_map=device if device == "cuda" else {"": device}, 
-                torch_dtype=torch.bfloat16,
+                device_map=device if device == "mps" or device == "cuda" else {"": device},
+                torch_dtype=torch.float16 if device == "mps" else torch.bfloat16,
                 max_memory=max_memory  # 添加 max_memory 参数
             ).eval()
 
@@ -148,7 +150,7 @@ def load_models(model_path, dtype, device="cuda", max_memory=None):
                 text_model = PeftModel.from_pretrained(
                     model=text_model, 
                     model_id=LORA_PATH, 
-                    device_map=device if device == "cuda" else {"": device},
+                    device_map=device if device == "mps" or device == "cuda" else {"": device},
                     quantization_config=nf4_config
                 )
                 text_model = text_model.merge_and_unload(
@@ -169,7 +171,12 @@ def load_models(model_path, dtype, device="cuda", max_memory=None):
             )
             image_adapter.eval().to(device)
         else:  # bf16
-            print("Loading in bfloat16")
+            if torch.backends.mps.is_available():
+                print("MPS Detected (Apple Silicon), fallback to fp16")
+                print("Loading in float16 (without AMP)")
+            else:
+                print("Loading in bfloat16")
+                
             print("Loading CLIP")
             clip_processor = AutoProcessor.from_pretrained(CLIP_PATH)
             clip_model = AutoModel.from_pretrained(CLIP_PATH).vision_model
@@ -188,8 +195,8 @@ def load_models(model_path, dtype, device="cuda", max_memory=None):
             print(f"Loading LLM: {model_path}")
             text_model = AutoModelForCausalLM.from_pretrained(
                 model_path, 
-                device_map="auto",
-                torch_dtype=torch.bfloat16,
+                device_map=device if device == "mps" or device == "cuda" else {"": device},
+                torch_dtype=torch.float16 if device == "mps" else torch.bfloat16,
                 max_memory=max_memory  # 添加 max_memory 参数
             ).eval()
 
@@ -198,7 +205,7 @@ def load_models(model_path, dtype, device="cuda", max_memory=None):
                 text_model = PeftModel.from_pretrained(
                     model=text_model, 
                     model_id=LORA_PATH, 
-                    device_map=device if device == "cuda" else {"": device}
+                    device_map=device if device == "mps" or device == "cuda" else {"": device},
                 )
                 text_model = text_model.merge_and_unload(
                     safe_merge=True
@@ -329,11 +336,19 @@ def stream_chat(input_images: List[Image.Image], caption_type: str, caption_leng
                 continue
 
             # Embed image
-            with torch.amp.autocast_mode.autocast(device, enabled=True):
+            if torch.backends.mps.is_available():
+                model.clip_model.to(torch.float16)
+                model.image_adapter.to(torch.float16)
+                pixel_values = pixel_values.to(torch.float16)
                 vision_outputs = model.clip_model(pixel_values=pixel_values, output_hidden_states=True)
                 image_features = vision_outputs.hidden_states
                 embedded_images = model.image_adapter(image_features).to(device)
-
+            else: 
+                with torch.amp.autocast_mode.autocast(device, enabled=True):
+                    vision_outputs = model.clip_model(pixel_values=pixel_values, output_hidden_states=True)
+                    image_features = vision_outputs.hidden_states
+                    embedded_images = model.image_adapter(image_features).to(device)
+                
             # Build the conversation
             convo = [
                 {
@@ -512,7 +527,7 @@ def joycaption2(
         llm_model_path_cache = os.path.join(comfy_model_dir, "cache--" + sanitized_model_name)
 
         # 初始设备设置为 'cuda'
-        device = 'cuda' if torch.cuda.is_available() else 'cpu'
+        device = 'cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu'
         model_loaded_on = device  # 跟踪模型加载在哪个设备上
 
         try:
@@ -757,7 +772,7 @@ def joycaption2_simple(
         llm_model_path_cache = os.path.join(comfy_model_dir, "cache--" + sanitized_model_name)
 
         # 初始设备设置为 'cuda'
-        device = 'cuda' if torch.cuda.is_available() else 'cpu'
+        device = 'cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu'
         model_loaded_on = device  # 跟踪模型加载在哪个设备上
 
         try: