Update app.py

app.py

@@ -15,6 +15,7 @@ import cv2
 
 from transformers import (
     Qwen2_5_VLForConditionalGeneration,
+    Gemma3ForConditionalGeneration,
     AutoProcessor,
     TextIteratorStreamer,
 )
@@ -27,10 +28,10 @@ MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
 
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
-# Load Cosmos-Reason1-7B
-MODEL_ID_M = "nvidia/Cosmos-Reason1-7B"
+# Load gemma-3-4b-it
+MODEL_ID_M = "google/gemma-3-4b-it"
 processor_m = AutoProcessor.from_pretrained(MODEL_ID_M, trust_remote_code=True)
-model_m = Qwen2_5_VLForConditionalGeneration.from_pretrained(
+model_m = Gemma3ForConditionalGeneration.from_pretrained(
     MODEL_ID_M,
     trust_remote_code=True,
     torch_dtype=torch.float16
@@ -85,7 +86,7 @@ def generate_image(model_name: str, text: str, image: Image.Image,
     """
     Generates responses using the selected model for image input.
     """
-    if model_name == "Cosmos-Reason1-7B":
+    if model_name == "gemma-3-4b-it":
         processor = processor_m
         model = model_m
     elif model_name == "docscopeOCR-7B-050425-exp":
@@ -138,7 +139,7 @@ def generate_video(model_name: str, text: str, video_path: str,
     """
     Generates responses using the selected model for video input.
     """
-    if model_name == "Cosmos-Reason1-7B":
+    if model_name == "gemma-3-4b-it":
         processor = processor_m
         model = model_m
     elif model_name == "docscopeOCR-7B-050425-exp":
@@ -244,9 +245,9 @@ with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
         with gr.Column():
             output = gr.Textbox(label="Output", interactive=False, lines=2, scale=2)
             model_choice = gr.Radio(
-                choices=["Cosmos-Reason1-7B", "docscopeOCR-7B-050425-exp", "Captioner-7B"],
+                choices=["gemma-3-4b-it", "docscopeOCR-7B-050425-exp", "Captioner-7B"],
                 label="Select Model",
-                value="Cosmos-Reason1-7B"
+                value="gemma-3-4b-it"
             )
             
             gr.Markdown("**Model Info**")