feat(provider): enable change onnx provider

wdtagger/__init__.py

@@ -2,6 +2,7 @@ import logging
 import os
 import time
 from collections import OrderedDict
+from typing import Any, Sequence
 
 import huggingface_hub
 import numpy as np
@@ -15,16 +16,9 @@ from rich.logging import RichHandler
 # Access console for rich text and logging
 console = rich.get_console()
 
-# Environment variables and file paths
-HF_TOKEN = os.environ.get(
-    "HF_TOKEN", ""
-)  # Token for authentication with HuggingFace API
-MODEL_FILENAME = "model.onnx"  # ONNX model filename
-LABEL_FILENAME = "selected_tags.csv"  # Labels CSV filename
-
-available_providers = rt.get_available_providers()
-supported_providers = ["CPUExecutionProvider", "CUDAExecutionProvider"]
-providers = list(set(available_providers) & set(supported_providers))
+HF_TOKEN = os.environ.get("HF_TOKEN", "")
+MODEL_FILENAME = "model.onnx"
+LABEL_FILENAME = "selected_tags.csv"
 
 
 def load_labels(dataframe) -> list[str]:
@@ -174,6 +168,7 @@ class Tagger:
         hf_token=HF_TOKEN,
         loglevel=logging.INFO,
         num_threads=None,
+        providers=None,
     ):
         """Initialize the Tagger object with the model repository and tokens.
 
@@ -184,16 +179,29 @@ class Tagger:
             loglevel (int, optional): Logging level. Defaults to logging.INFO.
             num_threads (int, optional): Number of threads for ONNX runtime. Defaults to None.
         """
+        if providers is None:
+            providers = ["CUDAExecutionProvider", "CPUExecutionProvider"]
         self.logger = logging.getLogger("wdtagger")
         self.logger.setLevel(loglevel)
         self.logger.addHandler(RichHandler())
         self.model_target_size = None
         self.cache_dir = cache_dir
         self.hf_token = hf_token
-        self.load_model(model_repo, cache_dir, hf_token, num_threads=num_threads)
+        self.load_model(
+            model_repo,
+            cache_dir,
+            hf_token,
+            num_threads=num_threads,
+            providers=providers,
+        )
 
     def load_model(
-        self, model_repo, cache_dir=None, hf_token=None, num_threads: int = None
+        self,
+        model_repo,
+        cache_dir=None,
+        hf_token=None,
+        num_threads: int = None,
+        providers: Sequence[str | tuple[str, dict[Any, Any]]] = None,
     ):
         """Load the model and tags from the specified repository.