Fix: Handle None LLAMA_CPP_MODEL_PATH and pre-download model

Dockerfile

@@ -5,7 +5,7 @@ ENV PIP_NO_CACHE_DIR=1
 ENV DOCKER_CONTAINER=true  
 ENV SPACE_APP_DATA=/data
 ENV HF_HOME=/data/huggingface
-ENV LLAMA_CPP_MODEL_PATH=/data/models/Hermes-2-Pro-Llama-3-8B-GGUF.Q4_K_M.gguf
+ENV LLAMA_CPP_MODEL_PATH=/data/models/Hermes-2-Pro-Llama-3-8B-Q4_K_M.gguf
 
 # Optimize llama-cpp-python build for CPU only
 ENV CMAKE_ARGS="-DLLAMA_BLAS=0 -DLLAMA_CUBLAS=0"
@@ -38,15 +38,15 @@ RUN python -m spacy download en_core_web_sm
 # Create directories that your app expects
 RUN mkdir -p /data/models /data/uploads /data/cache /data/logs /data/huggingface
 
-# Download GGUF model during build (BEFORE copying app code)
+# Download GGUF model during build
 RUN python -c "from huggingface_hub import hf_hub_download; \
     import shutil; \
     downloaded = hf_hub_download( \
         repo_id='NousResearch/Hermes-2-Pro-Llama-3-8B-GGUF', \
-        filename='Hermes-2-Pro-Llama-3-8B-GGUF.Q4_K_M.gguf', \
+        filename='Hermes-2-Pro-Llama-3-8B-Q4_K_M.gguf', \
         cache_dir='/data/huggingface' \
     ); \
-    shutil.copy(downloaded, '/data/models/Hermes-2-Pro-Llama-3-8B-GGUF.Q4_K_M.gguf')" && \
+    shutil.copy(downloaded, '/data/models/Hermes-2-Pro-Llama-3-8B-Q4_K_M.gguf')" && \
     echo "Model downloaded to /data/models/"
 
 # Copy app code

config/settings.py

@@ -48,7 +48,7 @@ class Settings(BaseSettings):
     LLAMA_CPP_ENABLED      : bool                                               = False  # Auto-enabled in HF Spaces
     LLAMA_CPP_MODEL_PATH   : Optional[Path]                                     = None   # Local path to GGUF model
     LLAMA_CPP_MODEL_REPO   : str                                                = "NousResearch/Hermes-2-Pro-Llama-3-8B-GGUF"
-    LLAMA_CPP_MODEL_FILE   : str                                                = "Hermes-2-Pro-Llama-3-8B-GGUF.Q4_K_M.gguf"
+    LLAMA_CPP_MODEL_FILE   : str                                                = "Hermes-2-Pro-Llama-3-8B-Q4_K_M.gguf"
     LLAMA_CPP_N_CTX        : int                                                = 4096   # Context window
     LLAMA_CPP_N_GPU_LAYERS : int                                                = -1     # -1 = all layers on GPU
     LLAMA_CPP_N_BATCH      : int                                                = 512    # Batch size for prompt processing