Update app.py

app.py

@@ -8,14 +8,13 @@ import sys
 import os
 from pathlib import Path
 from typing import Dict, Optional, Tuple
+import importlib
+import site
 
 # ============================================================================
 # CONFIGURATION
 # ============================================================================
 
-# We can now add all models. The installer will work for all of them.
-# The key is the lang code used by the UI.
-# The value is (Display Name, spaCy model name, type)
 MODEL_INFO: Dict[str, Tuple[str, str, str]] = {
     "de": ("German", "de_core_news_md", "spacy"),
     "en": ("English", "en_core_web_md", "spacy"),
@@ -23,10 +22,11 @@ MODEL_INFO: Dict[str, Tuple[str, str, str]] = {
     "grc-proiel-trf": ("Ancient Greek (PROIEL TRF)", "grc_proiel_trf", "grecy"),
     "grc-perseus-trf": ("Ancient Greek (Perseus TRF)", "grc_perseus_trf", "grecy"),
     "grc-proiel-lg": ("Ancient Greek (PROIEL LG)", "grc_proiel_lg", "grecy"),
-    # Add other models here as needed
+    "grc-perseus-lg": ("Ancient Greek (Perseus LG)", "grc_perseus_lg", "grecy"),
+    "grc-proiel-sm": ("Ancient Greek (PROIEL SM)", "grc_proiel_sm", "grecy"),
+    "grc-perseus-sm": ("Ancient Greek (Perseus SM)", "grc_perseus_sm", "grecy"),
 }
 
-# UI Text is unchanged
 UI_TEXT = {
     "de": {
         "title": "# 🔍 Mehrsprachiger Morpho-Syntaktischer Analysator",
@@ -77,25 +77,45 @@ UI_TEXT = {
         "tab_table": "Tabla",
         "tab_json": "JSON",
         "html_label": "Análisis de Dependencias",
-        "table_label": "Análisis de Dependencias",
+        "table_label": "Análisis Morfológico",
         "table_headers": ["Palabra", "Lema", "POS", "Etiqueta", "Morfología", "Dependencia"],
         "json_label": "Salida JSON",
         "error_message": "Error: "
     }
 }
 
-# This global dict holds the loaded models
+# This global dict holds the *loaded* models
 MODELS: Dict[str, Optional[spacy.Language]] = {}
 
 # ============================================================================
-# FIXED GRECY MODEL INSTALLATION
+# DEPENDENCY INSTALLATION
 # ============================================================================
 
+def install_spacy_transformers_once():
+    """
+    Installs spacy-transformers, required for all _trf models.
+    """
+    marker_file = Path(".spacy_transformers_installed")
+    if marker_file.exists():
+        print("✓ spacy-transformers already installed (marker found)")
+        return True
+    
+    print("Installing spacy-transformers (for _trf models)...")
+    cmd = [sys.executable, "-m", "pip", "install", "spacy-transformers"]
+    try:
+        subprocess.run(cmd, capture_output=True, text=True, check=True, timeout=300)
+        print("✓ Successfully installed spacy-transformers")
+        marker_file.touch()
+        return True
+    except Exception as e:
+        print(f"✗ FAILED to install spacy-transformers: {e}")
+        if hasattr(e, 'stderr'): print(e.stderr)
+        return False
+
 def install_grecy_model_from_github(model_name: str) -> bool:
     """
     Installs a greCy model from your specific GitHub Release.
-    This bypasses the broken 'grecy install' and uses --no-deps
-    to prevent dependency conflicts.
+    Uses --no-deps to prevent dependency conflicts.
     """
     marker_file = Path(f".{model_name}_installed")
     if marker_file.exists():
@@ -104,49 +124,37 @@ def install_grecy_model_from_github(model_name: str) -> bool:
 
     print(f"Installing grecy model: {model_name}...")
 
-    # --- Logic from our install_model.py script ---
+    # --- *** THIS IS THE CRITICAL FILENAME FIX *** ---
+    # The wheel filename's "distribution" part uses underscores,
+    # not hyphens, to be PEP 427 compliant.
     if model_name == "grc_proiel_trf":
         wheel_filename = "grc_proiel_trf-3.7.5-py3-none-any.whl"
-    elif model_name in ["grc_perseus_trf", "grc_proiel_lg", "grc_perseus_lg", "grc_proiel_sm", "grc_perseus_sm"]:
-        wheel_filename = f"{model_name.replace('_', '-')}-0.0.0-py3-none-any.whl"
     else:
-        print(f"✗ Unknown grecy model: {model_name}")
-        return False
-    # --- End logic ---
+        # All other models use the 0.0.0 version and underscore in name
+        # e.g., grc_perseus_trf-0.0.0-py3-none-any.whl
+        wheel_filename = f"{model_name}-0.0.0-py3-none-any.whl"
+    # --- *** END FIX *** ---
 
-    # This URL points to YOUR fork's release
     install_url = f"https://github.com/CrispStrobe/greCy/releases/download/v1.0-models/{wheel_filename}"
-
-    # This is the command that fixes everything
-    cmd = [
-        sys.executable, "-m", "pip", "install",
-        install_url,
-        "--no-deps"  # <-- The most important part!
-    ]
+    cmd = [sys.executable, "-m", "pip", "install", install_url, "--no-deps"]
 
     print(f"Running: {' '.join(cmd)}")
     try:
-        result = subprocess.run(
-            cmd,
-            capture_output=True,
-            text=True,
-            check=True,  # This will raise an error if pip fails
-            timeout=900  # 15 minute timeout
-        )
-        
-        if result.stdout:
-            print("STDOUT:", result.stdout)
-        if result.stderr:
-            print("STDERR:", result.stderr)
-            
+        result = subprocess.run(cmd, capture_output=True, text=True, check=True, timeout=900)
+        if result.stdout: print("STDOUT:", result.stdout)
+        if result.stderr: print("STDERR:", result.stderr)
         print(f"✓ Successfully installed {model_name} from GitHub")
         marker_file.touch()
         return True
-
     except subprocess.CalledProcessError as e:
         print(f"✗ Installation subprocess FAILED with code {e.returncode}")
         print("STDOUT:", e.stdout)
         print("STDERR:", e.stderr)
+        print("\n\n!! --- NOTE --- !!")
+        print(f"If the error is 'Invalid wheel filename', it means the files on your GitHub Release are named incorrectly.")
+        print(f"You must delete the release and re-run the fixed 'create_grecy_release.sh' script.")
+        print(f"The correct filename should be: '{wheel_filename}'.")
+        print("!! --- END NOTE --- !!\n\n")
         return False
     except Exception as e:
         print(f"✗ Installation exception: {e}")
@@ -154,14 +162,13 @@ def install_grecy_model_from_github(model_name: str) -> bool:
         return False
 
 # ============================================================================
-# MODEL LOADING (SIMPLIFIED)
+# MODEL LOADING (NOW WITH LAZY LOADING)
 # ============================================================================
 
 def load_spacy_model(model_name: str) -> Optional[spacy.Language]:
     """Load or install a standard spaCy model."""
     try:
-        nlp = spacy.load(model_name)
-        return nlp
+        return spacy.load(model_name)
     except OSError:
         print(f"Installing {model_name}...")
         try:
@@ -176,26 +183,14 @@ def load_grecy_model(model_name: str) -> Optional[spacy.Language]:
     Load a grecy model.
     First, installs it from our GitHub. Then, loads it.
     """
-    # Step 1: Ensure it's installed using our fixed method
-    install_success = install_grecy_model_from_github(model_name)
-    
-    if not install_success:
+    if not install_grecy_model_from_github(model_name):
         print(f"✗ Cannot load {model_name} because installation failed.")
         return None
-
-    # Step 2: Now that it's installed, just load it.
-    # All the complex strategies are no longer needed.
     try:
-        # We MUST refresh the import paths for spaCy to find the new package
         print("Refreshing importlib to find new package...")
-        import importlib
-        import site
         importlib.invalidate_caches()
-        try:
-            # Reload site-packages to find the newly pip-installed package
-            importlib.reload(site)
-        except Exception as e:
-            print(f"Warning: Could not reload 'site'. {e}")
+        try: importlib.reload(site)
+        except Exception: pass # Fails in some envs, but that's ok
 
         print(f"Trying: spacy.load('{model_name}')")
         nlp = spacy.load(model_name)
@@ -207,36 +202,43 @@ def load_grecy_model(model_name: str) -> Optional[spacy.Language]:
         traceback.print_exc()
         return None
 
-
 def initialize_models():
-    """Load all models at startup."""
+    """
+    Load all *non-grecy* models at startup for speed.
+    Grecy models will be lazy-loaded on first use.
+    Also ensures spacy-transformers is installed.
+    """
     print("\n" + "="*70)
     print("INITIALIZING MODELS")
     print("="*70 + "\n")
     
-    # We iterate over the key (lang_code) and the tuple (lang_name, model_name, type)
+    # First, ensure dependencies are met
+    install_spacy_transformers_once()
+    
+    loaded_count = 0
+    spacy_model_count = 0
+    
     for lang_code, (lang_name, model_name, model_type) in MODEL_INFO.items():
-        print(f"Loading {lang_name} ({model_name})...")
-        
-        if model_type == "grecy":
-            nlp = load_grecy_model(model_name)
-        else:
+        if model_type == "spacy":
+            spacy_model_count += 1
+            print(f"Loading {lang_name} ({model_name})...")
             nlp = load_spacy_model(model_name)
-        
-        # The key in the MODELS dict is the lang_code from MODEL_INFO
-        MODELS[lang_code] = nlp
-        
-        if nlp:
-            print(f"✓ {lang_name} ready\n")
+            MODELS[lang_code] = nlp # Store the loaded model
+            if nlp:
+                print(f"✓ {lang_name} ready\n")
+                loaded_count += 1
+            else:
+                print(f"✗ {lang_name} FAILED\n")
         else:
-            print(f"✗ {lang_name} FAILED\n")
+            # It's a grecy model, just mark as not loaded yet
+            print(f"✓ {lang_name} ({model_name}) will be loaded on first use.\n")
+            MODELS[lang_code] = None # Mark as available but not loaded
     
-    loaded = sum(1 for m in MODELS.values() if m is not None)
-    print(f"Loaded {loaded}/{len(MODELS)} models successfully")
+    print(f"Pre-loaded {loaded_count}/{spacy_model_count} standard models.")
     print("="*70 + "\n")
 
 # ============================================================================
-# ANALYSIS
+# ANALYSIS (WITH LAZY LOADING)
 # ============================================================================
 
 def get_analysis(ui_lang: str, model_lang_key: str, text: str):
@@ -249,18 +251,40 @@ def get_analysis(ui_lang: str, model_lang_key: str, text: str):
             return ([], [], "<p style='color: orange;'>No text provided.</p>",
                     gr.Button(value=ui_config["button_text"], interactive=True))
         
-        # model_lang_key is now the key from MODEL_INFO, e.g., "grc-proiel-trf"
-        if model_lang_key not in MODELS or MODELS[model_lang_key] is None:
-            raise ValueError(f"Model for {model_lang_key} is not available. Please check logs.")
+        # --- LAZY LOADING LOGIC ---
+        nlp = MODELS.get(model_lang_key)
+        
+        if nlp is None:
+            # Model hasn't been loaded yet
+            print(f"First use of {model_lang_key}. Loading model...")
+            
+            if model_lang_key not in MODEL_INFO:
+                 raise ValueError(f"Unknown model key: {model_lang_key}")
+            
+            _, model_name, model_type = MODEL_INFO[model_lang_key]
+            
+            if model_type == "grecy":
+                nlp = load_grecy_model(model_name)
+            else:
+                # This case should be pre-loaded, but as a fallback
+                nlp = load_spacy_model(model_name)
+            
+            if nlp is None:
+                # Failed to load
+                MODELS.pop(model_lang_key, None) # Remove from cache to try again next time
+                raise ValueError(f"Model for {model_lang_key} ({model_name}) FAILED to load. Check logs.")
+            else:
+                # Store in cache
+                MODELS[model_lang_key] = nlp
+                print(f"✓ {model_lang_key} is now loaded and cached.")
+        # --- END LAZY LOADING ---
         
-        nlp = MODELS[model_lang_key]
         doc = nlp(text)
         
         dataframe_output = []
         json_output = []
         
         for token in doc:
-            # FIXED: simplified lemma access for spaCy 3
             lemma_str = token.lemma_
             morph_str = str(token.morph) if token.has_morph() else ''
             dep_str = token.dep_ if doc.has_annotation("DEP") else ''
@@ -275,7 +299,6 @@ def get_analysis(ui_lang: str, model_lang_key: str, text: str):
             
             dataframe_output.append([token.text, lemma_str, pos_str, tag_str, morph_str, dep_str])
         
-        # Generate visualization
         html_out = ""
         if doc.has_annotation("DEP"):
             try:
@@ -324,7 +347,6 @@ def create_interface():
     config = UI_TEXT["en"]
     
     # Use the keys from MODEL_INFO as the choices
-    # The value passed to the backend will be "grc-proiel-trf"
     model_choices = list(MODEL_INFO.keys())
     
     with gr.Blocks(title="Multilingual Morpho-Syntactic Analyzer") as demo:
@@ -368,7 +390,6 @@ if __name__ == "__main__":
     print("MULTILINGUAL MORPHO-SYNTACTIC ANALYZER")
     print("="*70 + "\n")
     
-    # This will now use the new, robust installation
     initialize_models()
     
     demo = create_interface()