Update app.py

app.py

@@ -78,16 +78,7 @@ def txt_to_html(text):
     html_content += "</body></html>"
     return html_content
 
-@spaces.GPU(duration=80)
-def deidentify_doc(pdftext, maxtokens, temperature, top_probability):
-
-    llm = Llama(
-        model_path="models/Meta-Llama-3-8B-Instruct.Q8_0.gguf",
-        flash_attn=True,
-        n_gpu_layers=81,
-        n_batch=1024,
-        n_ctx=8192,
-    )
+def deidentify_doc(llm, pdftext, maxtokens, temperature, top_probability):
 
     #### Remove Dates ###
     prompt = "In the following text replace only the calendar dates with term [date]. Example: if input is 'Date of birth: 15/5/1959 calculated BP (Systolic 158.00 mm, Diastolic 124.95 mm)' output should be 'Date of birth: [date] calculated BP (Systolic 158.00 mm, Diastolic 124.95 mm)'"
@@ -185,8 +176,16 @@ def deidentify_doc(pdftext, maxtokens, temperature, top_probability):
         
     return output
 
+@spaces.GPU(duration=80)
 def pdf_to_text(files, maxtokens=2048, temperature=0, top_probability=0.95):
     files=[files]
+    llm = Llama(
+        model_path="models/Meta-Llama-3-8B-Instruct.Q8_0.gguf",
+        flash_attn=True,
+        n_gpu_layers=81,
+        n_batch=1024,
+        n_ctx=8192,
+    )
     for file in files:
         if not file:
             return 'Please provide a valid PDF'
@@ -197,7 +196,7 @@ def pdf_to_text(files, maxtokens=2048, temperature=0, top_probability=0.95):
             anonymized_text = ''
             for page_id in page2content:
                 pdftext = page2content[page_id]
-                anonymized_text += deidentify_doc(pdftext, maxtokens, temperature, top_probability)
+                anonymized_text += deidentify_doc(llm, pdftext, maxtokens, temperature, top_probability)
                 anonymized_text += '\n\n\n'
     return anonymized_text