Update app.py

app.py

@@ -20,7 +20,8 @@ START_TOKEN    = 128259
 NEW_BLOCK      = 128257
 EOS_TOKEN      = 128258
 AUDIO_BASE     = 128266
-AUDIO_IDS      = torch.arange(AUDIO_BASE, AUDIO_BASE + 4096)
+AUDIO_SPAN   = 4096 * 7                                # 28 672 Codes
+VALID_AUDIO  = torch.arange(AUDIO_BASE, AUDIO_BASE + AUDIO_SPAN)
 
 # 2) Logit‑Mask (NEW_BLOCK + Audio; EOS erst nach 1. Block) ----------
 class AudioMask(LogitsProcessor):
@@ -125,7 +126,7 @@ async def tts(ws: WebSocket):
             offset_len += len(new)
 
             # ----- weiter mit Cache (letzte PKV steht im Modell) ---------
-            past     = gen._past_key_values
+            past     = gen.past_key_values # Corrected cache access
             last_tok = new[-1]
 
             print("new tokens:", new[:25], flush=True)
@@ -147,6 +148,8 @@ async def tts(ws: WebSocket):
         pass
     except Exception as e:
         print("❌ WS‑Error:", e, flush=True)
+        import traceback
+        traceback.print_exc()
         if ws.client_state.name != "DISCONNECTED":
             await ws.close(code=1011)
     finally:
@@ -159,4 +162,4 @@ async def tts(ws: WebSocket):
 # 6) Dev‑Start --------------------------------------------------------
 if __name__ == "__main__":
     import uvicorn, sys
-    uvicorn.run("app:app", host="0.0.0.0", port=7860, log_level="info")
+    uvicorn.run("app:app", host="0.0.0.0", port=7860, log_level="info")