paige-ai/Virchow · model size mismatch for timm version==1.0.8

echoabc

Aug 15

hello,I'm running this code:

import timm
from timm.layers import SwiGLUPacked
import torch

how can i modify?thanks!

adamcasson

Paige AI org Aug 15

If you do:

timm.create_model("hf-hub:paige-ai/Virchow", pretrained=True, mlp_layer=SwiGLUPacked, act_layer=torch.nn.SiLU)

then the proper model args are parsed out of config.json automatically.

If you want to manually instantiate the model as in your code snippet, then you'll have to include the additional model_args from config.json:

timm.create_model(
    "vit_huge_patch14_224",
    pretrained=False,
    checkpoint_path=checkpoint_path,
    mlp_layer=SwiGLUPacked,
    act_layer=torch.nn.SiLU,
    img_size=224,
    init_values=1e-5,
    num_classes=0,
    mlp_ratio=5.3375,
    global_pool="",
    dynamic_img_size=True,
)

echoabc

Aug 16

It works. Thank you very much.

echoabc changed discussion status to closed Aug 16