Add config file and auto-download model feature to app.py

app.py

@@ -19,6 +19,29 @@ except ImportError:
 model = None
 rag_system = None
 
+def download_model_if_needed():
+    """Download LoRA adapter from Hugging Face Hub if not present."""
+    from huggingface_hub import snapshot_download
+
+    adapter_path = Path("./models/francis_botcon_lora")
+
+    if not adapter_path.exists():
+        print("Downloading LoRA adapter from Hugging Face Hub...")
+        try:
+            snapshot_download(
+                repo_id="rojaldo/francis-botcon-lora",
+                local_dir=str(adapter_path),
+                repo_type="model"
+            )
+            print("✓ LoRA adapter downloaded successfully")
+            return str(adapter_path)
+        except Exception as e:
+            print(f"⚠ Could not download model: {e}")
+            print("Using base model only")
+            return None
+
+    return str(adapter_path)
+
 def load_models():
     """Load models lazily on first use."""
     global model, rag_system
@@ -26,13 +49,8 @@ def load_models():
     if model is None:
         print("Initializing models...")
 
-        # Check for LoRA adapter
-        adapter_path = None
-        if Path("./models/francis_botcon_lora").exists():
-            adapter_path = "./models/francis_botcon_lora"
-            print("✓ LoRA adapter found")
-        else:
-            print("⚠ LoRA adapter not found, using base model")
+        # Check for or download LoRA adapter
+        adapter_path = download_model_if_needed()
 
         # Initialize model
         model = FrancisModel(adapter_path=adapter_path)

config/config.yaml

@@ -0,0 +1,65 @@
+# Francis Botcon Configuration
+
+# Model Configuration
+model:
+  base_model: "mistralai/Mistral-7B-Instruct-v0.2"  # Open-source high-quality 7B model
+  quantization: false
+  quantization_bits: 4
+  device: "cuda"  # Change to "cpu" if GPU not available
+  max_memory: 16  # GB
+
+# LoRA Configuration
+lora:
+  r: 16
+  lora_alpha: 32
+  target_modules: ["q_proj", "v_proj"]
+  lora_dropout: 0.05
+  bias: "none"
+  task_type: "CAUSAL_LM"
+
+# Training Configuration
+training:
+  epochs: 3
+  learning_rate: 2e-4
+  batch_size: 4
+  gradient_accumulation_steps: 2
+  max_seq_length: 1024
+  warmup_steps: 100
+  weight_decay: 0.01
+  output_dir: "./models/francis_botcon_lora"
+
+# Embeddings Configuration
+embeddings:
+  model: "sentence-transformers/all-MiniLM-L6-v2"
+  device: "cuda"
+
+# Vector DB Configuration
+vector_db:
+  type: "chromadb"  # Options: "chromadb" or "faiss"
+  db_path: "./data/vectordb"
+  chunk_size: 500
+  chunk_overlap: 100
+  top_k: 5
+  similarity_threshold: 0.6
+
+# Generation Configuration
+generation:
+  temperature: 0.7
+  max_tokens: 512
+  top_p: 0.9
+  top_k: 50
+  do_sample: true
+
+# Application Configuration
+app:
+  port: 7860
+  share: false
+  debug: true
+  log_level: "INFO"
+
+# Data paths
+data:
+  raw_dir: "./data/raw"
+  processed_dir: "./data/processed"
+  vectordb_dir: "./data/vectordb"
+  dataset_dir: "./data/dataset"

download_models.py

@@ -0,0 +1,28 @@
+#!/usr/bin/env python3
+"""Download models from Hugging Face Hub for the Space."""
+
+import os
+from pathlib import Path
+from huggingface_hub import snapshot_download
+
+def download_lora_adapter():
+    """Download the LoRA adapter from Hugging Face Hub."""
+    repo_id = "rojaldo/francis-botcon-lora"
+    cache_dir = Path("./models")
+    cache_dir.mkdir(parents=True, exist_ok=True)
+    
+    print(f"Downloading {repo_id}...")
+    try:
+        model_path = snapshot_download(
+            repo_id=repo_id,
+            local_dir=str(cache_dir / "francis_botcon_lora"),
+            repo_type="model"
+        )
+        print(f"✓ Model downloaded to: {model_path}")
+        return True
+    except Exception as e:
+        print(f"Error downloading model: {e}")
+        return False
+
+if __name__ == "__main__":
+    download_lora_adapter()