Update app.py

app.py

@@ -3,27 +3,35 @@ import os, gradio as gr, spaces
 from huggingface_hub import snapshot_download
 from indextts.infer import IndexTTS
 
-model_dir = snapshot_download("IndexTeam/IndexTTS-1.5", local_dir="checkpoints", local_dir_use_symlinks=False)
-cfg_path  = os.path.join(model_dir, "config.yaml")
+os.environ["DS_BUILD_OPS"] = "0"      # block any JIT build if deepspeed sneaks in
+
+model_dir = snapshot_download("IndexTeam/IndexTTS-1.5",
+                              local_dir="checkpoints",
+                              local_dir_use_symlinks=False)
+cfg_path = os.path.join(model_dir, "config.yaml")
 
 tts = None
 def load():
-	global tts
-	if tts is None: tts = IndexTTS(model_dir=model_dir, cfg_path=cfg_path)
+    global tts
+    if tts is None:
+        # the key bit → force plain pytorch
+        tts = IndexTTS(model_dir=model_dir,
+                       cfg_path=cfg_path,
+                       use_deepspeed=False)   # <- param exists in IndexTTS 1.5
 
 @spaces.GPU
 def synth(ref_wav, prompt):
-	load()
-	out = "out.wav"
-	tts.infer(ref_wav, prompt, out)
-	return out
+    load()
+    out = "out.wav"
+    tts.infer(ref_wav, prompt, out)
+    return out
 
 with gr.Blocks() as demo:
-	gr.Markdown("# index-tts 1.5 zerogpu")
-	txt  = gr.Textbox(label="text prompt")
-	ref  = gr.Audio(label="reference voice", type="filepath")
-	gen  = gr.Audio(label="generated speech", type="filepath")
-	gr.Button("generate").click(synth, [ref, txt], gen)
+    gr.Markdown("# index-tts 1.5 zerogpu")
+    txt = gr.Textbox(label="text prompt")
+    ref = gr.Audio(label="reference voice", type="filepath")
+    gen = gr.Audio(label="generated speech", type="filepath")
+    gr.Button("generate").click(synth, [ref, txt], gen)
 
 demo.queue()
 demo.launch(show_api=False, ssr_mode=False)