Update app.py

app.py

@@ -2,61 +2,45 @@ import gradio as gr
 import numpy as np
 import soundfile as sf
 import sentencepiece as spm
-from safetensors.numpy import load_file
+import torch
+from safetensors.torch import load_file
 
 # Load tokenizer
 sp = spm.SentencePieceProcessor()
 sp.load("tokenizer.model")
 
-# Load model tensors
+# Load model safely using torch (FIX)
 tensors = load_file("model.safetensors")
 
 print("Model loaded successfully")
 
+# Convert tensors to float32 safely
+weights = {}
 
-# Dequantization function
-def dequantize_weights(tensors):
+for name in tensors:
 
-    weights = {}
+    tensor = tensors[name]
 
-    for name in list(tensors.keys()):
+    if tensor.dtype == torch.bfloat16:
+        tensor = tensor.to(torch.float32)
 
-        if name.endswith("_scale"):
-            continue
+    weights[name] = tensor.cpu().numpy()
 
-        scale_name = name + "_scale"
+print("Weights converted successfully")
 
-        if scale_name in tensors:
 
-            weight_i8 = tensors[name].astype(np.float32)
-
-            scale = tensors[scale_name].astype(np.float32)
-
-            weights[name] = weight_i8 * scale
-
-        else:
-            weights[name] = tensors[name]
-
-    return weights
-
-
-weights = dequantize_weights(tensors)
-
-print("Weights dequantized successfully")
-
-
-# Text-to-Speech function
+# Text to speech function
 def text_to_speech(text):
 
     if text.strip() == "":
         return None
 
-    # Tokenize
+    # Tokenize text
     tokens = sp.encode(text)
 
     print("Tokens:", tokens)
 
-    # Generate dummy waveform (replace with real model inference)
+    # Generate simple waveform (placeholder)
     duration = max(2, len(tokens) * 0.1)
 
     sample_rate = 24000
@@ -65,7 +49,6 @@ def text_to_speech(text):
 
     audio = 0.2 * np.sin(2 * np.pi * 220 * t)
 
-    # Save file
     output_path = "output.wav"
 
     sf.write(output_path, audio, sample_rate)
@@ -73,27 +56,15 @@ def text_to_speech(text):
     return output_path
 
 
-# Gradio Interface
+# Gradio UI
 demo = gr.Interface(
-
     fn=text_to_speech,
-
-    inputs=gr.Textbox(
-        label="Enter text",
-        placeholder="Type text to convert into speech"
-    ),
-
-    outputs=gr.Audio(
-        label="Generated Speech",
-        type="filepath"
-    ),
-
-    title="Pocket TTS Text-to-Speech",
-
-    description="Enter text and generate speech using Pocket-TTS INT8 model"
+    inputs=gr.Textbox(label="Enter text"),
+    outputs=gr.Audio(type="filepath"),
+    title="Pocket TTS INT8 (CPU)",
+    description="Text-to-Speech using Pocket TTS model"
 )
 
-
 demo.launch(
     server_name="0.0.0.0",
     server_port=7860,