Error: torch not compiled with CUDA

[nahidalam]

I was following Roboflow notebook and got below error while predicting with the Grounding DINO model at inference.py

AssertionError: Torch not compiled with CUDA enabled

Looks like Grounding DINO is hardcoded to be inferenced with cuda so it is not able to do inference on the cpu

CONFIG_PATH = os.path.join(HOME, "GroundingDINO/groundingdino/config/GroundingDINO_SwinT_OGC.py")
WEIGHTS_NAME = "groundingdino_swint_ogc.pth"
WEIGHTS_PATH = os.path.join(HOME, "weights", WEIGHTS_NAME)

from groundingdino.util.inference import load_model, load_image, predict, annotate

model = load_model(CONFIG_PATH, WEIGHTS_PATH)

boxes, logits, phrases = predict(
    model=model, 
    image=image, 
    caption=TEXT_PROMPT, 
    box_threshold=BOX_TRESHOLD, 
    text_threshold=TEXT_TRESHOLD
)

Please suggest

[csv610]

Hi, Did you find any solution to this problem?

[bdubbs-clarifai]

@csv610 Theres a flag in the Model class that allows you to choose your device. Just set it to 'cpu' for ex: grounding_dino_model = Model(model_config_path=GROUNDING_DINO_CONFIG_PATH, model_checkpoint_path=GROUNDING_DINO_CHECKPOINT_PATH, device='cpu')

[csv610]

Hello, There are two methods "build_sam" and "build_model); One of them as given on some link is as follow from segment_anything import build_sam def load_model(model_config_path, model_checkpoint_path, cpu_only=False):    args = SLConfig.fromfile(model_config_path)    args.device = "cuda" if not cpu_only else "cpu"    model = build_model(args).to("cpu")    checkpoint = torch.load(model_checkpoint_path, map_location="cpu")    load_res = model.load_state_dict(clean_state_dict(checkpoint["model"]), strict=False)    print(load_res)    _ = model.eval()    return model

I had specified to(cpu) on the model but it did not work.

Thanks

On Wednesday, May 31, 2023 at 12:02:03 PM PDT, bdubbs-clarifai ***@***.***> wrote:  

@csv610 Theres a flag in the Model class that allows you to choose your device. Just set it to 'cpu' for ex: grounding_dino_model = Model(model_config_path=GROUNDING_DINO_CONFIG_PATH, model_checkpoint_path=GROUNDING_DINO_CHECKPOINT_PATH, device='cpu')

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

[tingxueronghua]

appalled to see this problem still not solved yet... Any plan on this?@SlongLiu

[ic]

appalled to see this problem still not solved yet... Any plan on this?@SlongLiu

The answer is what @bdubbs-clarifai reported a year ago.

More in detail, the Model class accepts a device argument, which defaults to cuda. Setting it to cpu works as expected.