Revert to CPU wheel by default, add audioop-lts to resolve Python 3.13 audioop removal, and default N_GPU_LAYERS to 0

app.py

@@ -86,7 +86,7 @@ SUPPORTED_MODELS = {
 CACHE_DIR = os.environ.get("HF_HOME", os.path.join(os.path.dirname(os.path.abspath(__file__)), ".cache"))
 N_CTX = int(os.environ.get("N_CTX", "4096"))
 NUM_THREADS = int(os.environ.get("NUM_THREADS", "2"))
-N_GPU_LAYERS = int(os.environ.get("N_GPU_LAYERS", "-1"))
+N_GPU_LAYERS = int(os.environ.get("N_GPU_LAYERS", "0"))
 
 os.makedirs(CACHE_DIR, exist_ok=True)
 

requirements.txt

@@ -1,4 +1,4 @@
---extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu121
+--extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu
 fastapi>=0.100.0
 uvicorn>=0.22.0
 llama-cpp-python>=0.2.75
@@ -7,5 +7,7 @@ sse-starlette>=1.6.0
 pydantic>=2.0.0
 requests>=2.31.0
 gradio>=4.0.0
+audioop-lts
+