Don't use tensorrt for embeddings

This commit is contained in:
Nicolas Mowen 2024-09-17 14:44:59 -06:00
parent 4b12e03ab1
commit 290b8b10e5

View File

@ -49,26 +49,24 @@ class Clip(OnnxClip):
@staticmethod @staticmethod
def _load_model(path: str, silent: bool): def _load_model(path: str, silent: bool):
providers = ort.get_available_providers() providers = []
options = [] options = []
for provider in providers: for provider in ort.get_available_providers():
if provider == "TensorrtExecutionProvider": if provider == "TensorrtExecutionProvider":
os.makedirs( continue
"/config/model_cache/tensorrt/ort/trt-engines", exist_ok=True elif provider == "OpenVINOExecutionProvider":
) # TODO need to verify openvino works correctly
os.makedirs("/config/model_cache/openvino/ort", exist_ok=True)
providers.append(provider)
options.append( options.append(
{ {
"trt_timing_cache_enable": True, "cache_dir": "/config/model_cache/openvino/ort",
"trt_timing_cache_path": "/config/model_cache/tensorrt/ort", "device_type": "GPU",
"trt_engine_cache_enable": True,
"trt_engine_cache_path": "/config/model_cache/tensorrt/ort/trt-engines",
} }
) )
elif provider == "OpenVINOExecutionProvider":
os.makedirs("/config/model_cache/openvino/ort", exist_ok=True)
options.append({"cache_dir": "/config/model_cache/openvino/ort"})
else: else:
providers.append(provider)
options.append({}) options.append({})
try: try: