Upload app.py with huggingface_hub

app.py

@@ -20,10 +20,7 @@ from maya3 import hub
 
 RELEASE = "maya-research/maya2-native"
 _paths = hub.resolve(RELEASE)  # downloads model shards + kyutai/mimi + speechbrain ECAPA
-BACKBONE_DIR = _paths["backbone_dir"]
-MIMI_DIR = _paths["mimi_dir"]
-ECAPA_DIR = _paths["ecapa_dir"]
-CKPT_DIR = _paths["ckpt"]
+CKPT_DIR = _paths["ckpt"]      # release dir; from_pretrained re-resolves offline-first
 
 import gradio as gr
 import numpy as np
@@ -55,13 +52,10 @@ def _get_tts():
 
         t0 = time.perf_counter()
         _tts = Maya3TTS.from_pretrained(
-            CKPT_DIR,
-            backbone_dir=BACKBONE_DIR,
-            mimi_dir=MIMI_DIR,
-            ecapa_dir=ECAPA_DIR,
+            CKPT_DIR,          # release dir; re-resolves (offline-first) + registers named voices
             device="cuda",
-            max_seats=1,   # smallest graph grid; demo serves one request at a time
-            capacity=1024,  # halves KV cache vs default; still ~80 s of audio per piece
+            max_seats=1,       # smallest graph grid; demo serves one request at a time
+            capacity=1024,     # halves KV cache vs default; still ~80 s of audio per piece
         )
         print(f"[maya2] engine ready in {time.perf_counter() - t0:.1f}s", flush=True)
     return _tts
@@ -161,7 +155,7 @@ with gr.Blocks(theme=gr.themes.Citrus(), css=CSS) as demo:
                     filterable=True,
                     info="Named voices ship with the model; or clone your own below.",
                 )
-            audio_out = gr.Audio(label="Generated speech", type="filepath", show_download_button=True)
+            audio_out = gr.Audio(label="Generated speech", type="filepath")
             btn = gr.Button("Generate speech", variant="primary")
         with gr.Column(scale=2):
             gr.Markdown("### Clone a voice (optional)")