Upload example_usage.py with huggingface_hub

example_usage.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
 """
-Exemple d'utilisation des modèles depuis Hugging Face
+Example usage of models from Hugging Face
 """
 
 import torch
@@ -10,34 +10,34 @@ from huggingface_hub import hf_hub_download
 import json
 import os
 
-# Import des modèles locaux (à adapter selon votre structure)
+# Import local models (to adapt to your structure)
 from color_model import ColorCLIP, SimpleTokenizer
 from hierarchy_model import Model as HierarchyModel, HierarchyExtractor
 from config import color_emb_dim, hierarchy_emb_dim
 
 def load_models_from_hf(repo_id: str, cache_dir: str = "./models_cache"):
     """
-    Charger les modèles depuis Hugging Face
+    Load models from Hugging Face
     
     Args:
-        repo_id: ID du repository Hugging Face
-        cache_dir: Dossier de cache local
+        repo_id: ID of the Hugging Face repository
+        cache_dir: Local cache directory
     """
     
     os.makedirs(cache_dir, exist_ok=True)
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     
-    print(f"📥 Chargement des modèles depuis '{repo_id}'...")
+    print(f"📥 Loading models from '{repo_id}'...")
     
-    # 1. Charger le modèle de couleur
-    print("  📦 Chargement du modèle de couleur...")
+    # 1. Loading color model
+    print("  📦 Loading color model...")
     color_model_path = hf_hub_download(
         repo_id=repo_id,
         filename="color_model.pt",
         cache_dir=cache_dir
     )
     
-    # Charger le vocabulaire
+    # Loading vocabulary
     vocab_path = hf_hub_download(
         repo_id=repo_id,
         filename="tokenizer_vocab.json",
@@ -56,10 +56,10 @@ def load_models_from_hf(repo_id: str, cache_dir: str = "./models_cache"):
     color_model.tokenizer = tokenizer
     color_model.load_state_dict(checkpoint)
     color_model.eval()
-    print("  ✅ Modèle de couleur chargé")
+    print("  ✅ Color model loaded")
     
-    # 2. Charger le modèle de hiérarchie
-    print("  📦 Chargement du modèle de hiérarchie...")
+    # 2. Loading hierarchy model
+    print("  📦 Loading hierarchy model...")
     hierarchy_model_path = hf_hub_download(
         repo_id=repo_id,
         filename="hierarchy_model.pth",
@@ -78,13 +78,13 @@ def load_models_from_hf(repo_id: str, cache_dir: str = "./models_cache"):
     hierarchy_extractor = HierarchyExtractor(hierarchy_classes, verbose=False)
     hierarchy_model.set_hierarchy_extractor(hierarchy_extractor)
     hierarchy_model.eval()
-    print("  ✅ Modèle de hiérarchie chargé")
+    print("  ✅ Hierarchy model loaded")
     
-    # 3. Charger le modèle principal CLIP
-    print("  📦 Chargement du modèle principal CLIP...")
+    # 3. Loading main CLIP model
+    print("  📦 Loading main CLIP model...")
     main_model_path = hf_hub_download(
         repo_id=repo_id,
-        filename="laion_explicable_model.pth",
+        filename="gap_clip.pth",
         cache_dir=cache_dir
     )
     
@@ -93,12 +93,12 @@ def load_models_from_hf(repo_id: str, cache_dir: str = "./models_cache"):
     )
     checkpoint = torch.load(main_model_path, map_location=device)
     
-    # Gérer différentes structures de checkpoint
+    # Handle different checkpoint structures
     if isinstance(checkpoint, dict):
         if 'model_state_dict' in checkpoint:
             clip_model.load_state_dict(checkpoint['model_state_dict'])
         else:
-            # Si le checkpoint est directement le state_dict
+            # If the checkpoint is directly the state_dict
             clip_model.load_state_dict(checkpoint)
     else:
         clip_model.load_state_dict(checkpoint)
@@ -107,9 +107,9 @@ def load_models_from_hf(repo_id: str, cache_dir: str = "./models_cache"):
     clip_model.eval()
     
     processor = CLIPProcessor.from_pretrained('laion/CLIP-ViT-B-32-laion2B-s34B-b79K')
-    print("  ✅ Modèle principal CLIP chargé")
+    print("  ✅ Main CLIP model loaded")
     
-    print("\n✅ Tous les modèles sont chargés!")
+    print("\n✅ All models loaded!")
     
     return {
         'color_model': color_model,
@@ -122,12 +122,12 @@ def load_models_from_hf(repo_id: str, cache_dir: str = "./models_cache"):
 
 def example_search(models, image_path: str = None, text_query: str = None):
     """
-    Exemple de recherche avec les modèles
+    Example search with the models
     
     Args:
-        models: Dictionnaire des modèles chargés
-        image_path: Chemin vers une image (optionnel)
-        text_query: Requête textuelle (optionnel)
+        models: Dictionary of loaded models
+        image_path: Path to an image (optional)
+        text_query: Text query (optional)
     """
     
     color_model = models['color_model']
@@ -136,17 +136,17 @@ def example_search(models, image_path: str = None, text_query: str = None):
     processor = models['processor']
     device = models['device']
     
-    print("\n🔍 Exemple de recherche...")
+    print("\n🔍 Example search...")
     
     if text_query:
-        print(f"  📝 Requête textuelle: '{text_query}'")
+        print(f"  📝 Text query: '{text_query}'")
         
         # Obtenir les embeddings de couleur et hiérarchie
         color_emb = color_model.get_text_embeddings([text_query])
         hierarchy_emb = hierarchy_model.get_text_embeddings([text_query])
         
-        print(f"  🎨 Embedding couleur: {color_emb.shape}")
-        print(f"  📂 Embedding hiérarchie: {hierarchy_emb.shape}")
+        print(f"  🎨 Color embedding: {color_emb.shape}")
+        print(f"  📂 Hierarchy embedding: {hierarchy_emb.shape}")
         
         # Obtenir les embeddings du modèle principal
         text_inputs = processor(text=[text_query], padding=True, return_tensors="pt")
@@ -156,13 +156,13 @@ def example_search(models, image_path: str = None, text_query: str = None):
             outputs = main_model(**text_inputs)
             text_features = outputs.text_embeds
         
-        print(f"  🎯 Embedding principal: {text_features.shape}")
+        print(f"  🎯 Main embedding: {text_features.shape}")
     
     if image_path and os.path.exists(image_path):
         print(f"  🖼️  Image: {image_path}")
         image = Image.open(image_path).convert("RGB")
         
-        # Obtenir les embeddings d'image
+        # Get image embeddings
         image_inputs = processor(images=[image], return_tensors="pt")
         image_inputs = {k: v.to(device) for k, v in image_inputs.items()}
         
@@ -170,37 +170,37 @@ def example_search(models, image_path: str = None, text_query: str = None):
             outputs = main_model(**image_inputs)
             image_features = outputs.image_embeds
         
-        print(f"  🎯 Embedding image: {image_features.shape}")
+        print(f"  🎯 Image embedding: {image_features.shape}")
 
 
 if __name__ == "__main__":
     import argparse
     
-    parser = argparse.ArgumentParser(description="Exemple d'utilisation des modèles")
+    parser = argparse.ArgumentParser(description="Example usage of models")
     parser.add_argument(
         "--repo-id",
         type=str,
         required=True,
-        help="ID du repository Hugging Face"
+        help="ID of the Hugging Face repository"
     )
     parser.add_argument(
         "--text",
         type=str,
         default="red dress",
-        help="Requête textuelle de recherche"
+        help="Text query for search"
     )
     parser.add_argument(
         "--image",
         type=str,
         default=None,
-        help="Chemin vers une image"
+        help="Path to an image"
     )
     
     args = parser.parse_args()
     
-    # Charger les modèles
+    # Load models
     models = load_models_from_hf(args.repo_id)
     
-    # Exemple de recherche
+    # Example search
     example_search(models, image_path=args.image, text_query=args.text)