unfortunately my guess was incorrect, and we are missing 13B with a minor RAM insufficiency. :(

app.py

@@ -92,7 +92,7 @@ You can learn more about LoRa here:
 
 This space is loading the model to RAM without performing any quantization, so the required RAM is high.
 
-You can merge models up to 13B. (If your adapter weights are too large, it might not work.)
+You can merge models up to 7B. (If your adapter weights are too large, it might not work.)
 """