don't stream text

app.py

@@ -7,8 +7,8 @@ from transformers import TextIteratorStreamer, AutoTokenizer, AutoModelForCausal
 from PIL import ImageDraw
 from torchvision.transforms.v2 import Resize
 
-# import subprocess
-# subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
+import subprocess
+subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
 
 model_id = "vikhyatk/moondream2"
 revision = "2024-05-20"
@@ -39,7 +39,11 @@ def answer_question(img, prompt):
     buffer = ""
     for new_text in streamer:
         buffer += new_text
-        yield buffer.strip()
+    
+    # Wait for the thread to finish
+    thread.join()
+
+    return buffer.strip()
 
 def extract_floats(text):
     # Regular expression to match an array of four floating point numbers
@@ -58,18 +62,18 @@ def extract_bbox(text):
         bbox = (x1, y1, x2, y2)
     return bbox
 
-def process_answer(img, answer):
-    if extract_bbox(answer) is not None:
-        x1, y1, x2, y2 = extract_bbox(answer)
-        draw_image = Resize(768)(img)
-        width, height = draw_image.size
-        x1, x2 = int(x1 * width), int(x2 * width)
-        y1, y2 = int(y1 * height), int(y2 * height)
-        bbox = (x1, y1, x2, y2)
-        ImageDraw.Draw(draw_image).rectangle(bbox, outline="red", width=3)
-        return gr.update(visible=True, value=draw_image)
+# def process_answer(img, answer):
+#     if extract_bbox(answer) is not None:
+#         x1, y1, x2, y2 = extract_bbox(answer)
+#         draw_image = Resize(768)(img)
+#         width, height = draw_image.size
+#         x1, x2 = int(x1 * width), int(x2 * width)
+#         y1, y2 = int(y1 * height), int(y2 * height)
+#         bbox = (x1, y1, x2, y2)
+#         ImageDraw.Draw(draw_image).rectangle(bbox, outline="red", width=3)
+#         return gr.update(visible=True, value=draw_image)
 
-    return gr.update(visible=False, value=None)
+#     return gr.update(visible=False, value=None)
 
 with gr.Blocks() as demo:
     gr.Markdown(
@@ -89,6 +93,6 @@ with gr.Blocks() as demo:
 
     submit.click(answer_question, [img, prompt], output)
     prompt.submit(answer_question, [img, prompt], output)
-    output.change(process_answer, [img, output], ann, show_progress=False)
+    # output.change(process_answer, [img, output], ann, show_progress=False)
 
 demo.queue().launch()