add args to use_eager

demo/inference_from_file.py

@@ -175,6 +175,11 @@ def parse_args():
         default=1.3,
         help="CFG (Classifier-Free Guidance) scale for generation (default: 1.3)",
     )
+    parser.add_argument(
+        "--use_eager",
+        action="store_true",
+        help="Use eager attention mode instead of flash_attention_2",
+    )
     
     return parser.parse_args()
 
@@ -244,11 +249,12 @@ def main():
     processor = VibeVoiceProcessor.from_pretrained(args.model_path)
 
     # Load model
+    attn_implementation = "flash_attention_2" if not args.use_eager else "eager"
     model = VibeVoiceForConditionalGenerationInference.from_pretrained(
         args.model_path,
         torch_dtype=torch.bfloat16,
         device_map='cuda',
-        attn_implementation="flash_attention_2" # we only test flash_attention_2
+        attn_implementation=attn_implementation # flash_attention_2 is recommended, eager may lead to lower audio quality
     )
 
     model.eval()

demo/text_examples/2p_short.txt

@@ -0,0 +1,2 @@
+Speaker 1: I heard there’s big news in TTS lately?
+Speaker 2: Yes! Microsoft Research just open-sourced VibeVoice. The model can generate speech up to 90 minutes long, with smooth delivery and rich emotion — it’s absolutely amazing.