space.gpu corrected

scripts/generate_scripts.py

@@ -6,10 +6,6 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
 import gradio as gr
 from dotenv import load_dotenv
 import spaces
-
-
-# Chargement du modèle et du tokenizer
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
@@ -18,8 +14,10 @@ model_id = "Qwen/Qwen2.5-0.5B"
 tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float32, trust_remote_code=True).to(device)
 
+
+@spaces.GPU()
 def generate_local(prompt: str, max_new_tokens: int = 350, temperature: float = 0.7) -> str:
-    device = model.device  # get the device the model is on
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # get the device the model is on
     inputs = tokenizer(prompt, return_tensors="pt").to(device)
     
     output_ids = model.generate(
@@ -32,7 +30,7 @@ def generate_local(prompt: str, max_new_tokens: int = 350, temperature: float =
     return tokenizer.decode(output_ids[0], skip_special_tokens=True)
 
 
-@spaces.GPU()
+
 def generate_script(prompt: str, word_count: int = 60) -> str:
     system_prompt = (
         "You are a professional video scriptwriter. "
@@ -41,7 +39,7 @@ def generate_script(prompt: str, word_count: int = 60) -> str:
     )
     return generate_local(system_prompt)
 
-@spaces.GPU()
+
 def one_word(query: str) -> str:
     prompt_final = (
         "Extract only the unique central theme of the following text in English in JSON format like this: "
@@ -56,7 +54,7 @@ def one_word(query: str) -> str:
         keyword = matches[0] if matches else ""
     return keyword.lower()
 
-@spaces.GPU()
+
 def generate_title(text: str) -> str:
     prompt_final = (
         "Generate a unique title for a YouTube Short video that is engaging and informative, "
@@ -64,7 +62,6 @@ def generate_title(text: str) -> str:
     )
     return generate_local(prompt_final, max_new_tokens=50, temperature=0.9).strip()
 
-@spaces.GPU()
 def generate_description(text: str) -> str:
     prompt_final = (
         "Write only the YouTube video description in English:\n"
@@ -75,7 +72,6 @@ def generate_description(text: str) -> str:
     )
     return generate_local(prompt_final, max_new_tokens=300, temperature=0.7).strip()
 
-@spaces.GPU()
 def generate_tags(text: str) -> list:
     prompt_final = (
         "List only the important keywords for this YouTube video, separated by commas, "

scripts/generate_subtitles.py

@@ -90,7 +90,7 @@ def transcribe_audio_to_subs(audio_path):
     des segments start/end/text, et sauvegarde en .srt.
     """
     print("🎙️ Transcription avec Whisper...")
-    model = whisper.load_model("medium", device="cuda" if torch.cuda.is_available() else "cpu")
+    model = whisper.load_model("medium")
     result = model.transcribe(audio_path)
 
     subtitles = [{