fix: optimization

app.py

@@ -1,11 +1,12 @@
 # Install poppler-utils
 import os
-import subprocess
 
 os.system("apt-get update")
 os.system("apt-get install --no-install-recommends -y poppler-utils tesseract-ocr")
 os.system("pip3 install --upgrade pip wheel setuptools")
-os.system("FLASH_ATTENTION_SKIP_CUDA_BUILD=true pip3 install flash-attn --no-build-isolation") 
+os.system(
+    "FLASH_ATTENTION_SKIP_CUDA_BUILD=true pip3 install flash-attn --no-build-isolation"
+)
 
 
 import time
@@ -392,7 +393,7 @@ with gr.Blocks(
         outputs=visualization_sub_tabs,
     )
 
-    demo.queue(default_concurrency_limit=1).launch(
+    demo.queue(default_concurrency_limit=4).launch(
         show_error=True,
         max_file_size="50mb",
     )

backends/smoldocling.py

@@ -10,7 +10,7 @@ from PIL import Image
 from vllm import LLM, SamplingParams
 
 # Configuration
-MAX_PAGES = 1
+MAX_PAGES = 2
 MODEL_PATH = "ds4sd/SmolDocling-256M-preview"
 PROMPT_TEXT = "Convert page to Docling."
 
@@ -21,7 +21,7 @@ chat_template = f"<|im_start|>User:<image>{PROMPT_TEXT}<end_of_utterance>\nAssis
 
 sampling_params = SamplingParams(
     temperature=0.0,
-    max_tokens=4096,
+    max_tokens=4096 * 2,
 )