Update routers/donut_inference.py

routers/donut_inference.py

@@ -2,46 +2,38 @@ import re
 import time
 import torch
 from transformers import DonutProcessor, VisionEncoderDecoderModel
+from config import settings
+from functools import lru_cache
 import os
 
-# Cache model and processor objects based on their path
-_model_cache = {}
 
-def load_model(processor_path: str, model_path: str):
-    key = (processor_path, model_path)
+@lru_cache(maxsize=1)
+def load_model():
+    processor = DonutProcessor.from_pretrained(settings.processor)
+    model = VisionEncoderDecoderModel.from_pretrained(settings.model)
 
-    if key not in _model_cache:
-        print(f"Loading new model: {model_path}")
-        processor = DonutProcessor.from_pretrained(processor_path)
-        model = VisionEncoderDecoderModel.from_pretrained(model_path)
-        device = "cuda" if torch.cuda.is_available() else "cpu"
-        model.to(device)
-        _model_cache[key] = (processor, model, device)
-    else:
-        print(f"Using cached model: {model_path}")
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    model.to(device)
 
-    return _model_cache[key]
+    return processor, model, device
 
 
-def process_document_donut(image, settings):
-    processor_path = settings.processor
-    model_path = settings.model
-
+def process_document_donut(image):
     worker_pid = os.getpid()
     print(f"Handling inference request with worker PID: {worker_pid}")
 
     start_time = time.time()
 
-    processor, model, device = load_model(processor_path, model_path)
+    processor, model, device = load_model()
 
-    # Prepare encoder inputs
+    # prepare encoder inputs
     pixel_values = processor(image, return_tensors="pt").pixel_values
 
-    # Prepare decoder inputs
+    # prepare decoder inputs
     task_prompt = "<s_cord-v2>"
     decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
 
-    # Generate output
+    # generate answer
     outputs = model.generate(
         pixel_values.to(device),
         decoder_input_ids=decoder_input_ids.to(device),
@@ -55,13 +47,14 @@ def process_document_donut(image, settings):
         return_dict_in_generate=True,
     )
 
-    # Postprocess
+    # postprocess
     sequence = processor.batch_decode(outputs.sequences)[0]
     sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
-    sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()
+    sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()  # remove first task start token
 
-    processing_time = time.time() - start_time
+    end_time = time.time()
+    processing_time = end_time - start_time
 
     print(f"Inference done, worker PID: {worker_pid}")
 
-    return processor.token2json(sequence), processing_time
+    return processor.token2json(sequence), processing_time