Update gemma_inference.py

gemma_inference.py

@@ -9,6 +9,7 @@ from typing import Generator, Optional
 import numpy as np
 from utils.snac_utils import generate_audio_data, get_snac
 from utils.vad import get_speech_timestamps, collect_chunks
+from huggingface_hub import login
 
 class GemmaOmniInference:
     """
@@ -20,15 +21,24 @@ class GemmaOmniInference:
         self.device = device
         self.model_id = model_id
         
+        # Authenticate with Hugging Face
+        hf_token = os.getenv("HF_TOKEN")
+        if hf_token:
+            print("Authenticating with Hugging Face...")
+            login(token=hf_token)
+        else:
+            print("Warning: HF_TOKEN not found. Make sure to set it in Space secrets.")
+        
         # Initialize models
         print("Loading Gemma 3n model...")
         self.model = Gemma3nForConditionalGeneration.from_pretrained(
             model_id, 
             device_map="auto", 
-            torch_dtype=torch.bfloat16
+            torch_dtype=torch.bfloat16,
+            token=hf_token
         ).eval()
         
-        self.processor = AutoProcessor.from_pretrained(model_id)
+        self.processor = AutoProcessor.from_pretrained(model_id, token=hf_token)
         
         # Keep the audio processing models
         print("Loading audio processing models...")