Refactor model loading to use direct model paths instead of snapshot downloads

app.py

@@ -1,37 +1,17 @@
-from pathlib import Path
-
 import gradio as gr
 import spaces
 import torch
-from huggingface_hub import snapshot_download
 from transformers import AutoProcessor, VoxtralForConditionalGeneration
 
-# Model paths and setup
-voxtral_mini_path = snapshot_download(
-    repo_id='mistralai/Voxtral-Mini-3B-2507',
-    revision='refs/pr/16',
-    local_dir=Path(__file__).parent / 'Voxtral-Mini-3B-2507',
-    resume_download=True,
-)
-print(f"Voxtral Mini model downloaded to: {voxtral_mini_path}")
-
-voxtral_small_path = snapshot_download(
-    repo_id='mistralai/Voxtral-Small-24B-2507',
-    revision='refs/pr/9',
-    local_dir=Path(__file__).parent / 'Voxtral-Small-24B-2507',
-    resume_download=True,
-)
-print(f"Voxtral Small model downloaded to: {voxtral_small_path}")
-
 device = "cuda" if torch.cuda.is_available() else "cpu"
 print(f"Using device: {device}")
 
 # Load model and processor
-voxtral_mini_processor = AutoProcessor.from_pretrained(voxtral_mini_path)
-voxtral_mini_model = VoxtralForConditionalGeneration.from_pretrained(voxtral_mini_path, torch_dtype=torch.bfloat16, device_map=device)
+voxtral_mini_processor = AutoProcessor.from_pretrained("MohamedRashad/Voxtral-Mini-3B-2507-transformers")
+voxtral_mini_model = VoxtralForConditionalGeneration.from_pretrained("MohamedRashad/Voxtral-Mini-3B-2507-transformers", torch_dtype=torch.bfloat16, device_map=device)
 
-voxtral_small_processor = AutoProcessor.from_pretrained(voxtral_small_path)
-voxtral_small_model = VoxtralForConditionalGeneration.from_pretrained(voxtral_small_path, torch_dtype=torch.bfloat16, device_map=device)
+voxtral_small_processor = AutoProcessor.from_pretrained("MohamedRashad/Voxtral-Small-24B-2507-transformers")
+voxtral_small_model = VoxtralForConditionalGeneration.from_pretrained("MohamedRashad/Voxtral-Small-24B-2507-transformers", torch_dtype=torch.bfloat16, device_map=device)
 
 @spaces.GPU()
 def process_audio(audio_path, model_name, language="en", max_tokens=500):
@@ -42,11 +22,11 @@ def process_audio(audio_path, model_name, language="en", max_tokens=500):
     if model_name == "Voxtral Mini (3B)":
         model = voxtral_mini_model
         processor = voxtral_mini_processor
-        repo_id = str(voxtral_mini_path)
+        repo_id = "MohamedRashad/Voxtral-Mini-3B-2507-transformers"
     elif model_name == "Voxtral Small (24B)":
         model = voxtral_small_model
         processor = voxtral_small_processor
-        repo_id = str(voxtral_small_path)
+        repo_id = "MohamedRashad/Voxtral-Small-24B-2507-transformers"
     else:
         return "Invalid model selected."