Update app.py

app.py

@@ -23,7 +23,7 @@ from transformers.image_utils import load_image
 # Constants for text generation
 MAX_MAX_NEW_TOKENS = 2048
 DEFAULT_MAX_NEW_TOKENS = 1024
-MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "8192"))
+MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
 
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
@@ -37,7 +37,7 @@ model_m = Qwen2_5_VLForConditionalGeneration.from_pretrained(
 ).to("cuda").eval()
 
 # Load MiMo-VL-7B-RL
-MODEL_ID_X = "XiaomiMiMo/MiMo-VL-7B-RL"
+MODEL_ID_X = "XiaomiMiMo/MiMo-VL-7B-SFT"
 processor_x = AutoProcessor.from_pretrained(MODEL_ID_X, trust_remote_code=True)
 model_x = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     MODEL_ID_X,
@@ -130,7 +130,7 @@ def generate_video(model_name: str, text: str, video_path: str,
     if model_name == "Cosmos-Reason1-7B":
         processor = processor_m
         model = model_m
-    elif model_name == "MiMo-VL-7B-RL":
+    elif model_name == "MiMo-VL-7B-SFT":
         processor = processor_x
         model = model_x
     else:
@@ -232,7 +232,7 @@ with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
         with gr.Column():
             output = gr.Textbox(label="Output", interactive=False)
             model_choice = gr.Dropdown(
-                choices=["Cosmos-Reason1-7B", "MiMo-VL-7B-RL"],
+                choices=["Cosmos-Reason1-7B", "MiMo-VL-7B-SFT"],
                 label="Select Model",
             value="Cosmos-Reason1-7B")