Upload 2 files

app.py

@@ -1,34 +1,60 @@
 import gradio as gr
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 import spacy
+import spaces
 
-# Load spaCy models
-nlp_en = spacy.load("en_core_web_sm")
-nlp_it = spacy.load("it_core_news_sm")
+# Placeholder for the GPU decorator
+def GPU(duration):
+    def decorator(func):
+        def wrapper(*args, **kwargs):
+            # Placeholder for GPU usage logic
+            print(f"Using GPU for {duration} seconds")
+            return func(*args, **kwargs)
+        return wrapper
+    return decorator
 
-# Load translation models and tokenizers
-tokenizer_en_it = AutoTokenizer.from_pretrained("LeonardPuettmann/Quadrifoglio-mt-en-it")
-model_en_it = AutoModelForSeq2SeqLM.from_pretrained("LeonardPuettmann/Quadrifoglio-mt-en-it")
+class ModelSingleton:
+    _instance = None
 
-tokenizer_it_en = AutoTokenizer.from_pretrained("LeonardPuettmann/Quadrifoglio-mt-it-en")
-model_it_en = AutoModelForSeq2SeqLM.from_pretrained("LeonardPuettmann/Quadrifoglio-mt-it-en")
+    def __new__(cls, *args, **kwargs):
+        if not cls._instance:
+            cls._instance = super(ModelSingleton, cls).__new__(cls, *args, **kwargs)
+        return cls._instance
 
+    def __init__(self):
+        if not hasattr(self, 'initialized'):
+            self.nlp_en = spacy.load("en_core_web_sm")
+            self.nlp_it = spacy.load("it_core_news_sm")
+
+            # Load translation models and tokenizers
+            self.tokenizer_en_it = AutoTokenizer.from_pretrained("LeonardPuettmann/Quadrifoglio-mt-en-it")
+            self.model_en_it = AutoModelForSeq2SeqLM.from_pretrained("LeonardPuettmann/Quadrifoglio-mt-en-it")
+
+            self.tokenizer_it_en = AutoTokenizer.from_pretrained("LeonardPuettmann/Quadrifoglio-mt-it-en")
+            self.model_it_en = AutoModelForSeq2SeqLM.from_pretrained("LeonardPuettmann/Quadrifoglio-mt-it-en")
+
+            self.initialized = True
+
+model_singleton = ModelSingleton()
+
+@spaces.GPU(duration=120)
 def generate_response_en_it(input_text):
-    input_ids = tokenizer_en_it("translate English to Italian: " + input_text, return_tensors="pt").input_ids
-    output = model_en_it.generate(input_ids, max_new_tokens=256)
-    return tokenizer_en_it.decode(output[0], skip_special_tokens=True)
+    input_ids = model_singleton.tokenizer_en_it("translate English to Italian: " + input_text, return_tensors="pt").input_ids
+    output = model_singleton.model_en_it.generate(input_ids, max_new_tokens=256)
+    return model_singleton.tokenizer_en_it.decode(output[0], skip_special_tokens=True)
 
+@spaces.GPU(duration=120)
 def generate_response_it_en(input_text):
-    input_ids = tokenizer_it_en("translate Italian to English: " + input_text, return_tensors="pt").input_ids
-    output = model_it_en.generate(input_ids, max_new_tokens=256)
-    return tokenizer_it_en.decode(output[0], skip_special_tokens=True)
+    input_ids = model_singleton.tokenizer_it_en("translate Italian to English: " + input_text, return_tensors="pt").input_ids
+    output = model_singleton.model_it_en.generate(input_ids, max_new_tokens=256)
+    return model_singleton.tokenizer_it_en.decode(output[0], skip_special_tokens=True)
 
 def translate_text(input_text, direction):
     if direction == "en-it":
-        nlp = nlp_en
+        nlp = model_singleton.nlp_en
         generate_response = generate_response_en_it
     elif direction == "it-en":
-        nlp = nlp_it
+        nlp = model_singleton.nlp_it
         generate_response = generate_response_it_en
     else:
         return "Invalid direction selected."

requirements.txt

@@ -1,6 +1,7 @@
 gradio
 transformers
 spacy
+spaces
 torch
 en-core-web-sm @ https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.7.1/en_core_web_sm-3.7.1-py3-none-any.whl
 it-core-news-sm @ https://github.com/explosion/spacy-models/releases/download/it_core_news_sm-3.8.0/it_core_news_sm-3.8.0-py3-none-any.whl