Updating Side-By-Side Space

app.py

@@ -4,6 +4,26 @@ import time
 import sys
 from datetime import datetime
 
+# Import model configuration
+try:
+    from model_paths import (
+        BASE_MODEL_REPO,
+        NOVEL_MODEL_REPO,
+        BASE_MODEL_FILENAME,
+        NOVEL_MODEL_FILENAME,
+        MODELS_DIR
+    )
+except ImportError:
+    # Default values if import fails
+    BASE_MODEL_REPO = "unsloth/Qwen3-4B-GGUF"
+    NOVEL_MODEL_REPO = "mradermacher/Clinician-Note-2.0a-i1-GGUF"
+    BASE_MODEL_FILENAME = "Qwen3-4B-Q4_K_M.gguf"
+    NOVEL_MODEL_FILENAME = "Clinician-Note-2.0a.i1-Q4_K_M.gguf"
+    MODELS_DIR = "models"
+
+    # Create models directory
+    os.makedirs(MODELS_DIR, exist_ok=True)
+
 # Try to import llama_cpp
 try:
     from llama_cpp import Llama
@@ -14,27 +34,110 @@ except ImportError:
     LLAMA_CPP_AVAILABLE = False
     print("llama_cpp is not available. Running in fallback mode.")
 
-# Define the GGUF model paths
-BASE_MODEL_PATH = "unsloth/Qwen3-4B-GGUF/Qwen3-4B-Q4_K_M.gguf"
-NOVEL_MODEL_PATH = "mradermacher/Clinician-Note-2.0a-i1-GGUF/Clinician-Note-2.0a.i1-Q4_K_M.gguf"
+# Set up initial model paths
+BASE_MODEL_PATH = os.path.join(MODELS_DIR, BASE_MODEL_FILENAME)
+NOVEL_MODEL_PATH = os.path.join(MODELS_DIR, NOVEL_MODEL_FILENAME)
 
 # Initialize models
 base_model = None
 novel_model = None
 
 
+def download_model(model_repo_id, model_filename, progress=None, progress_start=0.0, progress_end=0.1):
+    """Download a model from Hugging Face Hub if it doesn't exist locally"""
+    try:
+        from huggingface_hub import hf_hub_download
+        import os
+
+        # Create models directory if it doesn't exist
+        os.makedirs("models", exist_ok=True)
+
+        # Define the local path for the model
+        local_path = os.path.join("models", model_filename)
+
+        # Check if model already exists locally
+        if os.path.exists(local_path):
+            print(f"Model {model_filename} already exists at {local_path}")
+            return local_path
+
+        # If progress is provided, update it
+        if progress is not None:
+            progress(progress_start, desc=f"Downloading {model_filename}... This may take a while")
+
+        print(f"Downloading {model_filename} from Hugging Face Hub...")
+
+        # Download the model from Hugging Face
+        downloaded_path = hf_hub_download(
+            repo_id=model_repo_id,
+            filename=model_filename,
+            local_dir="models",
+            local_dir_use_symlinks=False
+        )
+
+        # If progress is provided, update it
+        if progress is not None:
+            progress(progress_end, desc=f"Downloaded {model_filename}")
+
+        print(f"Model downloaded to {downloaded_path}")
+        return downloaded_path
+
+    except Exception as e:
+        print(f"Error downloading model {model_filename}: {str(e)}")
+        if progress is not None:
+            progress(progress_end, desc=f"Error downloading model: {str(e)}")
+        return None
+
+
 def load_models(progress=None):
     """Load the llama.cpp models"""
-    global base_model, novel_model
+    global base_model, novel_model, BASE_MODEL_PATH, NOVEL_MODEL_PATH
 
     if not LLAMA_CPP_AVAILABLE:
         print("llama_cpp not available, cannot load models")
         return False
 
     try:
+        # Download base model if needed
+        base_model_repo = "unsloth/Qwen3-4B-GGUF"
+        base_model_filename = "Qwen3-4B-Q4_K_M.gguf"
+
+        if progress is not None:
+            progress(0.1, desc="Checking for base model...")
+
+        base_model_path = download_model(
+            base_model_repo,
+            base_model_filename,
+            progress,
+            0.1, 0.3
+        )
+
+        if not base_model_path:
+            raise Exception(f"Failed to download base model {base_model_filename}")
+
+        BASE_MODEL_PATH = base_model_path
+
+        # Download novel model if needed
+        novel_model_repo = "mradermacher/Clinician-Note-2.0a-i1-GGUF"
+        novel_model_filename = "Clinician-Note-2.0a.i1-Q4_K_M.gguf"
+
+        if progress is not None:
+            progress(0.4, desc="Checking for novel model...")
+
+        novel_model_path = download_model(
+            novel_model_repo,
+            novel_model_filename,
+            progress,
+            0.4, 0.6
+        )
+
+        if not novel_model_path:
+            raise Exception(f"Failed to download novel model {novel_model_filename}")
+
+        NOVEL_MODEL_PATH = novel_model_path
+
         # Load base model
         if progress is not None:
-            progress(0.2, desc="Loading base model... This may take a few minutes")
+            progress(0.7, desc="Loading base model... This may take a few minutes")
 
         print(f"Loading base model from {BASE_MODEL_PATH}")
         base_model = Llama(
@@ -45,7 +148,7 @@ def load_models(progress=None):
 
         # Load novel model
         if progress is not None:
-            progress(0.7, desc="Loading novel model... This may take a few minutes")
+            progress(0.9, desc="Loading novel model... This may take a few minutes")
 
         print(f"Loading novel model from {NOVEL_MODEL_PATH}")
         novel_model = Llama(

model_paths.py

@@ -0,0 +1,12 @@
+# Model configuration file
+
+# Hugging Face Hub repository IDs
+BASE_MODEL_REPO = "unsloth/Qwen3-4B-GGUF"
+NOVEL_MODEL_REPO = "mradermacher/Clinician-Note-2.0a-i1-GGUF"
+
+# Model filenames
+BASE_MODEL_FILENAME = "Qwen3-4B-Q4_K_M.gguf"
+NOVEL_MODEL_FILENAME = "Clinician-Note-2.0a.i1-Q4_K_M.gguf"
+
+# Local directory for models
+MODELS_DIR = "models"

requirements.txt

@@ -5,4 +5,6 @@ torch==2.7.0
 accelerate==1.6.0
 sentencepiece==0.2.0
 protobuf==6.30.2
-llama-cpp-python==0.3.9
+llama-cpp-python==0.3.9
+requests>=2.31.0
+tqdm>=4.65.0