uploaded models

gguf_engine.py

@@ -17,20 +17,28 @@ from llama_cpp.llama_chat_format import Llava15ChatHandler
 # PATHS — update these to match your setup
 # ==========================================
 import os
-# This dynamically finds the folder right next to your app.py script
-BASE_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), "gguf_models")
+from huggingface_hub import hf_hub_download
 
-LLM_PATH = f"{BASE_DIR}/medgemma_q4km.gguf"
+# ==========================================
+# DYNAMIC MODEL DOWNLOADER (Bridge to Model Repo)
+# ==========================================
+MODEL_REPO = "shrishSVaidya/VAJRAM-Models" # Change this if you named it differently
+
+print("Fetching models from Hugging Face Hub (this takes a moment on first boot)...")
+
+# hf_hub_download pulls the file into the Space's local cache. 
+# If it's already downloaded, it loads instantly in 0.01 seconds!
+LLM_PATH = hf_hub_download(repo_id=MODEL_REPO, filename="gguf_models/medgemma_q4km.gguf")
 
 VISION_MMPROJ_PATHS = {
-    "module3": f"{BASE_DIR}/medgemma_Bone_marrow_vision.gguf",
-    "base":    f"{BASE_DIR}/medgemma_vision_base.gguf",
+    "module3": hf_hub_download(repo_id=MODEL_REPO, filename="gguf_models/medgemma_Bone_marrow_vision.gguf"),
+    "base":    hf_hub_download(repo_id=MODEL_REPO, filename="gguf_models/medgemma_vision_base.gguf"),
 }
 
 LLM_LORA_PATHS = {
-    "module2": f"{BASE_DIR}/lora_module2.gguf",
-    "module3": f"{BASE_DIR}/lora_module3.gguf",
-    "module4": f"{BASE_DIR}/lora_module4.gguf",
+    "module2": hf_hub_download(repo_id=MODEL_REPO, filename="gguf_models/lora_module2.gguf"),
+    "module3": hf_hub_download(repo_id=MODEL_REPO, filename="gguf_models/lora_module3.gguf"),
+    "module4": hf_hub_download(repo_id=MODEL_REPO, filename="gguf_models/lora_module4.gguf"),
     "default": None,
 }
 

requirements.txt

@@ -6,4 +6,5 @@ langchain-huggingface
 sentence-transformers
 faiss-cpu
 Pillow
-matplotlib
+matplotlib
+huggingface_hub