tweak 4

app.py

@@ -3,26 +3,48 @@ import gradio as gr
 import copy
 from llama_cpp import Llama
 from huggingface_hub import hf_hub_download  
+from huggingface_hub.file_download import http_get
 
 # huggingface-cli download microsoft/Phi-3-mini-4k-instruct-gguf Phi-3-mini-4k-instruct-q4.gguf --local-dir .
 # huggingface-cli download LoneStriker/OpenBioLLM-Llama3-8B-GGUF --local-dir ./llama3-gguf
 # Explicitly create the cache directory if it doesn't exist
-cache_dir = os.path.expanduser("~/.cache/huggingface/hub")
-os.makedirs(cache_dir, exist_ok=True)
 
-# Use the cache directory for huggingface downloads
-os.environ["HF_HOME"] = cache_dir
 
-llm = Llama(
-    # model_path="./Phi-3-mini-4k-instruct-q4.gguf",
-    # model_path="./llama3-gguf/OpenBioLLM-Llama3-8B-Q5_K_M.gguf",
-    model_path = hf_hub_download(
-        repo_id=os.environ.get("REPO_ID", "LoneStriker/OpenBioLLM-Llama3-8B-GGUF"),
-        filename=os.environ.get("MODEL_FILE", "Llama3-8B-Q5_K_M.gguf"),
-    ),
-    n_ctx=2048,
-    n_gpu_layers=50, # change n_gpu_layers if you have more or less VRAM 
-) 
+def load_model(
+    directory: str = ".",
+    model_name: str = "OpenBioLLM-Llama3-8B-Q5_K_M.gguf",
+    model_url: str = "https://huggingface.co/LoneStriker/OpenBioLLM-Llama3-8B-GGUF/resolve/main/OpenBioLLM-Llama3-8B-Q5_K_M.gguf"
+):
+    final_model_path = os.path.join(directory, model_name)
+    
+    print("Downloading all files...")
+    if not os.path.exists(final_model_path):
+        with open(final_model_path, "wb") as f:
+            http_get(model_url, f)
+    os.chmod(final_model_path, 0o777)
+    print("Files downloaded!")
+    
+    model = Llama(
+        model_path=final_model_path,
+        n_ctx=2000,
+        n_parts=1,
+    )
+    
+    print("Model loaded!")
+    return model
+
+# llm = Llama(
+#     # model_path="./Phi-3-mini-4k-instruct-q4.gguf",
+#     # model_path="./llama3-gguf/OpenBioLLM-Llama3-8B-Q5_K_M.gguf",
+#     model_path = hf_hub_download(
+#         repo_id=os.environ.get("REPO_ID", "LoneStriker/OpenBioLLM-Llama3-8B-GGUF"),
+#         filename=os.environ.get("MODEL_FILE", "Llama3-8B-Q5_K_M.gguf"),
+#     ),
+#     n_ctx=2048,
+#     n_gpu_layers=50, # change n_gpu_layers if you have more or less VRAM 
+# ) 
+
+llm = load_model()
 
 # print("here")
 def generate_text(