Fixed

requirements.txt

@@ -1,10 +1,5 @@
-# Core deps
 gradio==4.44.1
 huggingface_hub==0.25.2
-
-# FIXED: Use pre-built wheel + CMake for fast HF build
-llama-cpp-python-cpu==0.3.1
-cmake>=3.28.0  # Required for 0.3+ builds
-
-# Pin to avoid schema bugs
-gradio-client==0.17.0
+gradio-client==0.17.0
+# Direct working wheel — builds in <60 seconds
+https://github.com/abetlen/llama-cpp-python/releases/download/v0.2.89/llama_cpp_python-0.2.89-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl