feat: try fix...

app.py

@@ -112,18 +112,18 @@ LANGUAGES = {
 SYSTEM_PROMPT = "You are a speech synthesizer that generates natural, realistic, and human-like conversational audio from dialogue text."
 MODEL_PATH = "fnlp/MOSS-TTSD-v0"
 SPT_CONFIG_PATH = "XY_Tokenizer/config/xy_tokenizer_config.yaml"
-SPT_CHECKPOINT_PATH = "XY_Tokenizer/weights/xy_tokenizer.ckpt"
+# SPT_CHECKPOINT_PATH = "XY_Tokenizer/weights/xy_tokenizer.ckpt"
 MAX_CHANNELS = 8
 
 from huggingface_hub import hf_hub_download
 
-ckpt_path = hf_hub_download(
+SPT_CHECKPOINT_PATH = hf_hub_download(
     repo_id="fnlp/XY_Tokenizer_TTSD_V0",
     filename="xy_tokenizer.ckpt",
     cache_dir="XY_Tokenizer/weights"
 )
 
-print("Checkpoint downloaded to:", ckpt_path)
+print("Checkpoint downloaded to:", SPT_CHECKPOINT_PATH)
 
 # Global variables for caching loaded models
 tokenizer = None
@@ -137,7 +137,8 @@ def initialize_model():
     
     if tokenizer is None:
         print("Initializing model...")
-        device = "cuda" if torch.cuda.is_available() else "cpu"
+        device = "cuda"
+        print(f"Using {device}")
         tokenizer, model, spt = load_model(MODEL_PATH, SPT_CONFIG_PATH, SPT_CHECKPOINT_PATH)
         spt = spt.to(device)
         model = model.to(device)

requirements.txt

@@ -1,4 +1,4 @@
-torch>=2.0.0
+torch==2.5.1
 torchaudio>=2.0.0
 transformers>=4.30.0
 gradio>=4.0.0
@@ -14,4 +14,4 @@ openai
 PyYAML
 einops
 huggingface_hub
-https://github.com/Dao-AILab/flash-attention/releases/download/v2.8.0.post2/flash_attn-2.8.0.post2+cu12torch2.7cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
+https://github.com/Dao-AILab/flash-attention/releases/download/v2.8.0.post2/flash_attn-2.8.0.post2+cu12torch2.5cxx11abiTRUE-cp310-cp310-linux_x86_64.whl