[tokenizer] Handles import huggingface model zoo exception case

extensions/tokenizers/src/main/python/huggingface_converter.py

@@ -41,9 +41,20 @@ def save_model(self, model_info, args: Namespace, temp_dir: str):
         if not os.path.exists(temp_dir):
             os.makedirs(temp_dir)
 
-        hf_pipeline = self.load_model(model_id)
-        # Save tokenizer.json to temp dir
-        self.save_tokenizer(hf_pipeline, temp_dir)
+        try:
+            hf_pipeline = self.load_model(model_id)
+        except Exception as e:
+            logging.warning(f"Failed to load model: {model_id}.")
+            logging.warning(e, exc_info=True)
+            return False, "Failed to load model", -1
+
+        try:
+            # Save tokenizer.json to temp dir
+            self.save_tokenizer(hf_pipeline, temp_dir)
+        except Exception as e:
+            logging.warning(f"Failed to save tokenizer: {model_id}.")
+            logging.warning(e, exc_info=True)
+            return False, "Failed to save tokenizer", -1
 
         # Save config.json just for reference
         config = hf_hub_download(repo_id=model_id, filename="config.json")
@@ -112,7 +123,7 @@ def jit_trace_model(self, hf_pipeline, model_id: str, temp_dir: str,
             logging.info(f"Saving torchscript model: {model_name}.pt ...")
             model_file = os.path.join(temp_dir, f"{model_name}.pt")
             script_module.save(model_file)
-        except (RuntimeError, ValueError) as e:
+        except Exception as e:
             logging.warning(f"Failed to trace model: {model_id}.")
             logging.warning(e, exc_info=True)
             return None

extensions/tokenizers/src/main/python/huggingface_models.py

@@ -56,23 +56,29 @@ def __init__(self, output_dir: str):
         self.temp_dir = f"{self.output_dir}/tmp"
 
     def list_models(self, args: Namespace) -> List[dict]:
+        import_all = os.environ.get("HF_IMPORT_ALL")
+
         api = HfApi()
         if args.model_name:
             models = api.list_models(filter="pytorch",
                                      search=args.model_name,
                                      sort="downloads",
                                      direction=-1,
                                      limit=args.limit)
-            if not models:
-                logging.warning(f"no model found: {args.model_name}.")
+            import_all = True
         else:
             models = api.list_models(filter=f"{args.category},pytorch",
                                      sort="downloads",
                                      direction=-1,
                                      limit=args.limit)
-            if not models:
+        if not models:
+            if args.model_name:
+                logging.warning(f"no model found: {args.model_name}.")
+            else:
                 logging.warning(f"no model matches category: {args.category}.")
 
+            return []
+
         ret = []
         for model_info in models:
             model_id = model_info.modelId
@@ -83,7 +89,7 @@ def list_models(self, args: Namespace) -> List[dict]:
                 continue
 
             languages = get_lang_tags(model_info)
-            if "en" not in languages and not os.environ["HF_IMPORT_ALL"]:
+            if "en" not in languages and not import_all:
                 logging.warning(f"Skip non-English model: {model_id}.")
                 continue
 
@@ -94,6 +100,12 @@ def list_models(self, args: Namespace) -> List[dict]:
                     logging.info(f"Skip converted model: {model_id}.")
                     continue
 
+            if model_info.downloads < 50 and not import_all:
+                logging.info(
+                    f"Skip model {model_info.modelId}, downloads {model_info.downloads} < 50"
+                )
+                continue
+
             try:
                 config = hf_hub_download(repo_id=model_id,
                                          filename="config.json")

extensions/tokenizers/src/main/python/model_zoo_importer.py

@@ -49,9 +49,17 @@ def main():
         model_info = model["model_info"]
         converter = SUPPORTED_TASK[task]
 
-        result, reason, size = converter.save_model(model_info, args, temp_dir)
-        if not result:
-            logging.error(f"{model_info.modelId}: {reason}")
+        try:
+            result, reason, size = converter.save_model(
+                model_info, args, temp_dir)
+            if not result:
+                logging.error(f"{model_info.modelId}: {reason}")
+        except Exception as e:
+            logging.warning(f"Failed to convert model: {model_info.modelId}.")
+            logging.warning(e, exc_info=True)
+            result = False
+            reason = "Failed to convert model"
+            size = -1
 
         huggingface_models.update_progress(model_info, converter.application,
                                            result, reason, size, args.cpu_only)