ui beautify

app.py

@@ -109,17 +109,23 @@ def inference(img):
             use_cache=False # Cache helps speed on CPU significantly
         )
 
-        # Threading is REQUIRED for streaming
-        # The model generates in a separate thread, while the main thread
-        # yields from the streamer iterator.
+        logger.info("Starting model generation...")
         thread = Thread(target=model.generate, kwargs=generation_kwargs)
         thread.start()
 
         generated_text = ""
+        status_indicator = " **[...Processing]** ⏳"
         for new_text in streamer:
-            generated_text += new_text
+            if generated_text == "":
+                # Todo: remove the removal of first character when the model is fixed
+                generated_text = new_text[1:] 
+            else:
+                generated_text += new_text
             # Yielding here updates the Gradio textbox in real-time
-            yield generated_text
+            logger.info("yielding: " + generated_text)
+            yield generated_text + status_indicator
+        
+        yield generated_text
 
     except Exception as e:
         import traceback
@@ -138,10 +144,13 @@ description = '''
 This model represents a **world first in tech access and linguistic rights**, specifically trained to recognize 
 the diverse character and glyph repertoire of Mexico's 68 indigenous languages.
 
-**How to use:** Simply upload an image containing text in any Mexican indigenous language.
-**Note:** Running on CPU. Streaming is enabled so you can see progress immediately.
+**How to use:** Simply upload an image containing text in any Mexican indigenous language, and the model will 
+detect and recognize the text.
+
+### Warning: as this free demonstrator space uses only CPU, even small image could take up to 5 minutes, so be patient.
 
-🔗 [PaddleOCR Documentation](https://github.com/PaddlePaddle/PaddleOCR)
+🔗 [TachiwinOCR Model](https://huggingface.co/tachiwin/PaddleOCR-VL-Tachiwin-BF16)
+🔗 [TachiwinOCR Training Code](https://github.com/ljcamargo/tachiwin_paddleocrvl_finetuning)
 '''
 
 examples = [
@@ -166,11 +175,14 @@ example_labels = """
 | lac.jpg | Lacandon | wa quin chen u'yicob a t'ʌnex, wa yʌn in wu'yicob a ba' cu ya'aric C'uj? Tin t'ʌn, mʌ' in wu'yicob a t'ʌnex, yʌn in wu'yicob a ba' cu ya'aric C'uj. Yʌn in man in wa'aricob a ba' caj in wirajob yejer a ba' caj in wu'yajob ―baxuc tu ya'araj Pedro ti' u jach ts'urirob. Jeroj tune', chich t'ʌn Pedro yejer Juan ten u jach ts'urirob u winiquirob judío, caj ts'oquij caj cha'b u binob ten u jach ts'urirob. |
 """
 
-css = ".output_image, .input_image {height: 40rem !important; width: 100% !important;} .output_markdown {min-height: 30rem !important;}"
-
-# Note: We replaced gr.Interface with gr.Blocks or used the generator compatible interface
-# But standard Interface supports generators in newer Gradio versions.
-# Just ensuring concurrency_limit is set.
+css = """
+.output_image, .input_image {height: 40rem !important; width: 100% !important;} 
+.output_markdown {min-height: 30rem !important;}
+.output_markdown p, .output_markdown {
+    font-size: 1.3rem !important; 
+    line-height: 1.6 !important;
+}
+"""
 
 demo = gr.Interface(
     fn=inference,
@@ -208,4 +220,4 @@ demo = gr.Interface(
 )
 
 if __name__ == "__main__":
-    demo.queue().launch()
+    demo.queue().launch(debug=True)